Post by thread:[PIC]: PIC32 for camera capture

SUPPOSING that the PIC can DMA transfer data from the sensor to an external memory in one cycle (it can't), the maximum frame rate you could accomplish would be 16fps at 80MHz, and you would have no time to process it. Usually, image processing is much more intensive than the capture (that is, you will need to access each pixel several times). Your external memory needs to be at least 10MiB (16MiB is the logical choice) for color images (YCbCr) or 5MiB (8MiB being the logical choice) for gray scale images, just to hold one frame. Do you really need such a large image? We do fingerprint recognition with VGA sensors (640x480) cropped to 440x380. We use a dsPIC to pre-process the image but the capture is done with a Xilinx CPLD coupled to a 256KiBx8 static RAM. We can capture approx. 13 frames per second and can determine if a valid fingerprint is present in less than 1/2 second. Best regards, Isaac Em 21/5/2012 18:13, Andre Abelian escreveu: {Quote hidden}> Hi all, > > does any one know if pic32 will have enough speed to capture a camera data? > I am thinking to use 5 mp camera directly connected to pic and of course there is > a RAM too for a buffer. main purpose of the camera is to sense a motion change and > send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA > part for some how I believe PIC32 may do the job. once the motion is detected we can stop it > send it out then continue. I do not need to do continues live video transmit or some thing. > > any idea or suggestion will appreciate > > thanks > > Andre

Hi Jason, Finally I chose the part MT9P401 from Aptina. based on my understanding the pixel output period is every 10.4nS. so the pic should be able to capture 12 bit "12 pins" data ever 10 nS and save it in SDRAM. basically there are 5.000.000 pixels I need to capture and store them in SDRAM. how about pic24EP 120mhz is it faster then pic32? in my case I really need 16 bit. pic32 is 80mhz/2 40 mhz and pic24EP is 120mhz/2 60 mhz. what do you think? any suggestion will appreciate thanks Andre. ________________________________ From: Jason White <.....whitewaterssoftwareinfoKILLspam@spam@gmail.com> To: Microcontroller discussion list - Public. <piclistKILLspammit.edu> Sent: Monday, May 21, 2012 2:25 PM Subject: Re: [PIC]: PIC32 for camera capture On Mon, May 21, 2012 at 5:13 PM, Andre Abelian <.....abelian.andreKILLspam.....yahoo.com> wrote: {Quote hidden}> Hi all, > > does any one know if pic32 will have enough speed to capture a camera data? > I am thinking to use 5 mp camera directly connected to pic and of course there is > a RAM too for a buffer. main purpose of the camera is to sense a motion change and > send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA > part for some how I believe PIC32 may do the job. once the motion is detected we can stop it > send it out then continue. I do not need to do continues live video transmit or some thing. > > any idea or suggestion will appreciate I am no expert at this, but the frame rate you want to read from the camera and process the data will determine whether or not your processor is fast enough. More details please ! -- Jason White

Hi Isaac, I didn't see your replay for some reason. the purpose of this camera is to capture a person and send email as attachment. the quality is important. in your case how pic is communicating with CPLD? in my case the only part I am not clear is I have to sense a motion. if old picture and new are not same then I have to send new picture out thru wireless to pc. what about PIC24EP? thanks Andre ________________________________ From: Isaac Marino Bavaresco <EraseMEisaacbavarescospam_OUTTakeThisOuTyahoo.com.br> To: Microcontroller discussion list - Public. <piclistspam_OUTmit.edu> Sent: Monday, May 21, 2012 3:14 PM Subject: Re: [PIC]: PIC32 for camera capture SUPPOSING that the PIC can DMA transfer data from the sensor to an external memory in one cycle (it can't), the maximum frame rate you could accomplish would be 16fps at 80MHz, and you would have no time to process it. Usually, image processing is much more intensive than the capture (that is, you will need to access each pixel several times). Your external memory needs to be at least 10MiB (16MiB is the logical choice) for color images (YCbCr) or 5MiB (8MiB being the logical choice) for gray scale images, just to hold one frame. Do you really need such a large image? We do fingerprint recognition with VGA sensors (640x480) cropped to 440x380. We use a dsPIC to pre-process the image but the capture is done with a Xilinx CPLD coupled to a 256KiBx8 static RAM. We can capture approx. 13 frames per second and can determine if a valid fingerprint is present in less than 1/2 second. Best regards, Isaac Em 21/5/2012 18:13, Andre Abelian escreveu: {Quote hidden}> Hi all, > > does any one know if pic32 will have enough speed to capture a camera data? > I am thinking to use 5 mp camera directly connected to pic and of course there is > a RAM too for a buffer. main purpose of the camera is to sense a motion change and > send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA > part for some how I believe PIC32 may do the job. once the motion is detected we can stop it > send it out then continue. I do not need to do continues live video transmit or some thing. > > any idea or suggestion will appreciate > > thanks > > Andre

Em 22/5/2012 23:47, Andre Abelian escreveu: > Hi Isaac, > > I didn't see your replay for some reason. the purpose of this camera is to capture > a person and send email as attachment. the quality is important. in your case how pic is > communicating with CPLD? I implemented a bit-banged 8-bit bus. It can transfer 8MB/s. > in my case the only part I am not clear is I have to sense a motion. You need to store at least two images (10MB for B/W images, 20MB for color images) and compare them. The biggest problem I see is that you cannot simply bytewise compare the images, you must use some intelligent algorithm to detect real movement, not noise or small changes in lighting. Prepare to scan both images several times (read each pixel more than once) and do a lot of math (FFT, filtering, wavelet, etc.). I think that this task is up to a much larger processor (ARM9 @ 400MIPS, etc). > if old picture and new are not same They will never be the same, always there are a lot of noise, moving shadows, difference in lighting, vibration, etc. > then I have to send new picture out thru > wireless to pc. what about PIC24EP? They can do 60MIPS. Just to scan two 5MP images and compare them would take 0.25s *IF* you could access each pixel in one cycle and do the compare in one cycle. As far as I know, no PIC or dsPIC have external bus interface that can access such a large memory (not even they have SDRAM controllers), and surely not at one transfer per clock. If you implement your own bus, it will be much slower. My implementation takes 5 instructions to do a transfer and it is optimized as hell, and the address generation is done by the CPLD, The dsPIC just needs to mange the data and control lines. As I said, you need a large MPU (100's of MIPS) with some MB of SDRAM and a large cache. Best regards, Isaac {Quote hidden}> > thanks > > Andre > > > > > ________________________________ > From: Isaac Marino Bavaresco <@spam@isaacbavarescoKILLspamyahoo.com.br> > To: Microcontroller discussion list - Public. <KILLspampiclistKILLspammit.edu> > Sent: Monday, May 21, 2012 3:14 PM > Subject: Re: [PIC]: PIC32 for camera capture > > SUPPOSING that the PIC can DMA transfer data from the sensor to an > external memory in one cycle (it can't), the maximum frame rate you > could accomplish would be 16fps at 80MHz, and you would have no time to > process it. > Usually, image processing is much more intensive than the capture (that > is, you will need to access each pixel several times). > > Your external memory needs to be at least 10MiB (16MiB is the logical > choice) for color images (YCbCr) or 5MiB (8MiB being the logical choice) > for gray scale images, just to hold one frame. > > Do you really need such a large image? We do fingerprint recognition > with VGA sensors (640x480) cropped to 440x380. > We use a dsPIC to pre-process the image but the capture is done with a > Xilinx CPLD coupled to a 256KiBx8 static RAM. > We can capture approx. 13 frames per second and can determine if a valid > fingerprint is present in less than 1/2 second. > > > Best regards, > > Isaac > > > > Em 21/5/2012 18:13, Andre Abelian escreveu: >> Hi all, >> >> does any one know if pic32 will have enough speed to capture a camera data? >> I am thinking to use 5 mp camera directly connected to pic and of course there is >> a RAM too for a buffer. main purpose of the camera is to sense a motion change and >> send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA >> part for some how I believe PIC32 may do the job. once the motion is detected we can stop it >> send it out then continue. I do not need to do continues live video transmit or some thing. >> >> any idea or suggestion will appreciate >> >> thanks >> >> Andre

I don't think PIC is your way to go in this case. You might be able to detect the motion in grayscale precisely, but face recognition, person identification from image etc can not be done real-time via PIC. As others stated you can check out ARM9-10-11. Why don't you switch to FPGA ? An entry level FPGA will be capable of doing these kinds of tasks very well without that much of an effort. On Wed, May 23, 2012 at 1:48 PM, Isaac Marino Bavaresco <RemoveMEisaacbavarescoTakeThisOuTyahoo.com.br> wrote: {Quote hidden}> Em 22/5/2012 23:47, Andre Abelian escreveu: >> Hi Isaac, >> >> I didn't see your replay for some reason. the purpose of this camera is to capture >> a person and send email as attachment. the quality is important. in your case how pic is >> communicating with CPLD? > > I implemented a bit-banged 8-bit bus. It can transfer 8MB/s. > > >> in my case the only part I am not clear is I have to sense a motion. > > > You need to store at least two images (10MB for B/W images, 20MB for > color images) and compare them. > The biggest problem I see is that you cannot simply bytewise compare the > images, you must use some intelligent algorithm to detect real movement, > not noise or small changes in lighting. > Prepare to scan both images several times (read each pixel more than > once) and do a lot of math (FFT, filtering, wavelet, etc.). > I think that this task is up to a much larger processor (ARM9 @ 400MIPS, > etc). > > >> if old picture and new are not same > > > They will never be the same, always there are a lot of noise, moving > shadows, difference in lighting, vibration, etc. > > >> then I have to send new picture out thru >> wireless to pc. what about PIC24EP? > > > They can do 60MIPS. Just to scan two 5MP images and compare them would > take 0.25s *IF* you could access each pixel in one cycle and do the > compare in one cycle. > As far as I know, no PIC or dsPIC have external bus interface that can > access such a large memory (not even they have SDRAM controllers), and > surely not at one transfer per clock. > If you implement your own bus, it will be much slower. My implementation > takes 5 instructions to do a transfer and it is optimized as hell, and > the address generation is done by the CPLD, The dsPIC just needs to > mange the data and control lines. > > > As I said, you need a large MPU (100's of MIPS) with some MB of SDRAM > and a large cache. > > > Best regards, > > Isaac > > >> >> thanks >> >> Andre >> >> >> >> >> ________________________________ >> From: Isaac Marino Bavaresco <spamBeGoneisaacbavarescospamBeGoneyahoo.com.br> >> To: Microcontroller discussion list - Public. <TakeThisOuTpiclistEraseMEspam_OUTmit.edu> >> Sent: Monday, May 21, 2012 3:14 PM >> Subject: Re: [PIC]: PIC32 for camera capture >> >> SUPPOSING that the PIC can DMA transfer data from the sensor to an >> external memory in one cycle (it can't), the maximum frame rate you >> could accomplish would be 16fps at 80MHz, and you would have no time to >> process it. >> Usually, image processing is much more intensive than the capture (that >> is, you will need to access each pixel several times). >> >> Your external memory needs to be at least 10MiB (16MiB is the logical >> choice) for color images (YCbCr) or 5MiB (8MiB being the logical choice) >> for gray scale images, just to hold one frame. >> >> Do you really need such a large image? We do fingerprint recognition >> with VGA sensors (640x480) cropped to 440x380. >> We use a dsPIC to pre-process the image but the capture is done with a >> Xilinx CPLD coupled to a 256KiBx8 static RAM. >> We can capture approx. 13 frames per second and can determine if a valid >> fingerprint is present in less than 1/2 second. >> >> >> Best regards, >> >> Isaac >> >> >> >> Em 21/5/2012 18:13, Andre Abelian escreveu: >>> Hi all, >>> >>> does any one know if pic32 will have enough speed to capture a camera data? >>> I am thinking to use 5 mp camera directly connected to pic and of course there is >>> a RAM too for a buffer. main purpose of the camera is to sense a motion change and >>> send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA >>> part for some how I believe PIC32 may do the job. once the motion is detected we can stop it >>> send it out then continue. I do not need to do continues live video transmit or some thing. >>> >>> any idea or suggestion will appreciate >>> >>> thanks >>> >>> Andre > >

Did you notice that 10ns is 100MHz? That means that you have to read one 12-bit pixel from the sensor and store it to external memory faster than any PIC can execute one instruction.. Even a 400MHz ARM would have a hard time doing this without hardware assistance. I tried and desisted. It is doable, but the ARM becomes enslaved by the image capture. Isaac Em 22/5/2012 23:33, Andre Abelian escreveu: {Quote hidden}> Hi Jason, > > Finally I chose the part MT9P401 from Aptina. based on my understanding > the pixel output period is every 10.4nS. so the pic should be able to capture 12 bit "12 pins" data ever > 10 nS and save it in SDRAM. basically there are 5.000.000 pixels I need to capture and store them in SDRAM. > how about pic24EP 120mhz is it faster then pic32? in my case I really need 16 bit. > pic32 is 80mhz/2 40 mhz and pic24EP is 120mhz/2 60 mhz. what do you think? > > any suggestion will appreciate > > thanks > > Andre. > > > ________________________________ > From: Jason White <RemoveMEwhitewaterssoftwareinfoTakeThisOuTgmail.com> > To: Microcontroller discussion list - Public. <piclistEraseME.....mit.edu> > Sent: Monday, May 21, 2012 2:25 PM > Subject: Re: [PIC]: PIC32 for camera capture > > On Mon, May 21, 2012 at 5:13 PM, Andre Abelian <EraseMEabelian.andreyahoo.com> wrote: >> Hi all, >> >> does any one know if pic32 will have enough speed to capture a camera data? >> I am thinking to use 5 mp camera directly connected to pic and of course there is >> a RAM too for a buffer. main purpose of the camera is to sense a motion change and >> send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA >> part for some how I believe PIC32 may do the job. once the motion is detected we can stop it >> send it out then continue. I do not need to do continues live video transmit or some thing. >> >> any idea or suggestion will appreciate > I am no expert at this, but the frame rate you want to read from the > camera and process the data will determine whether or not your > processor is fast enough. More details please ! > >

Isaac, I agree. My question is let say I use cpld to capture the data now i need to compare portion of the image to determine as motion change. I am thinking to use dual ported ram to capture the image and use pic to access to ram and compare data? what do you think? Andre ________________________________ From: Isaac Marino Bavaresco <RemoveMEisaacbavarescoEraseMEEraseMEyahoo.com.br> To: Microcontroller discussion list - Public. <RemoveMEpiclistspam_OUTKILLspammit.edu> Sent: Wednesday, May 23, 2012 5:56 AM Subject: Re: [PIC]: PIC32 for camera capture Did you notice that 10ns is 100MHz? That means that you have to read one 12-bit pixel from the sensor and store it to external memory faster than any PIC can execute one instruction.. Even a 400MHz ARM would have a hard time doing this without hardware assistance. I tried and desisted. It is doable, but the ARM becomes enslaved by the image capture. Isaac Em 22/5/2012 23:33, Andre Abelian escreveu: {Quote hidden}> Hi Jason, > > Finally I chose the part MT9P401 from Aptina. based on my understanding > the pixel output period is every 10.4nS. so the pic should be able to capture 12 bit "12 pins" data ever > 10 nS and save it in SDRAM. basically there are 5.000.000 pixels I need to capture and store them in SDRAM. > how about pic24EP 120mhz is it faster then pic32? in my case I really need 16 bit. > pic32 is 80mhz/2 40 mhz and pic24EP is 120mhz/2 60 mhz. what do you think? > > any suggestion will appreciate > > thanks > > Andre. > > > ________________________________ > From: Jason White <RemoveMEwhitewaterssoftwareinfoTakeThisOuTspamgmail.com> > To: Microcontroller discussion list - Public. <EraseMEpiclistspamspamBeGonemit.edu> > Sent: Monday, May 21, 2012 2:25 PM > Subject: Re: [PIC]: PIC32 for camera capture > > On Mon, May 21, 2012 at 5:13 PM, Andre Abelian <RemoveMEabelian.andreKILLspamyahoo.com> wrote: >> Hi all, >> >> does any one know if pic32 will have enough speed to capture a camera data? >> I am thinking to use 5 mp camera directly connected to pic and of course there is >> a RAM too for a buffer. main purpose of the camera is to sense a motion change and >> send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA >> part for some how I believe PIC32 may do the job. once the motion is detected we can stop it >> send it out then continue. I do not need to do continues live video transmit or some thing. >> >> any idea or suggestion will appreciate > I am no expert at this, but the frame rate you want to read from the > camera and process the data will determine whether or not your > processor is fast enough. More details please ! > >

Yigit, the way I am seeing iti need to use cpld to capture the image in background and use pic to access to ram using dual ported ram that way motion can be detected like you suggested using gray scale portion. Now my question is for pic do I need ram? or I can use internal program memory instead? thanks AA ________________________________ From: Yigit Turgut <spamBeGoney.turgutSTOPspamEraseMEgmail.com> To: Microcontroller discussion list - Public. <KILLspampiclistspamBeGonemit.edu> Sent: Wednesday, May 23, 2012 4:25 AM Subject: Re: [PIC]: PIC32 for camera capture I don't think PIC is your way to go in this case. You might be able to detect the motion in grayscale precisely, but face recognition, person identification from image etc can not be done real-time via PIC. As others stated you can check out ARM9-10-11. Why don't you switch to FPGA ? An entry level FPGA will be capable of doing these kinds of tasks very well without that much of an effort. On Wed, May 23, 2012 at 1:48 PM, Isaac Marino Bavaresco <EraseMEisaacbavarescoEraseMEyahoo.com.br> wrote: {Quote hidden}> Em 22/5/2012 23:47, Andre Abelian escreveu: >> Hi Isaac, >> >> I didn't see your replay for some reason. the purpose of this camera is to capture >> a person and send email as attachment. the quality is important. in your case how pic is >> communicating with CPLD? > > I implemented a bit-banged 8-bit bus. It can transfer 8MB/s. > > >> in my case the only part I am not clear is I have to sense a motion. > > > You need to store at least two images (10MB for B/W images, 20MB for > color images) and compare them. > The biggest problem I see is that you cannot simply bytewise compare the > images, you must use some intelligent algorithm to detect real movement, > not noise or small changes in lighting. > Prepare to scan both images several times (read each pixel more than > once) and do a lot of math (FFT, filtering, wavelet, etc.). > I think that this task is up to a much larger processor (ARM9 @ 400MIPS, > etc). > > >> if old picture and new are not same > > > They will never be the same, always there are a lot of noise, moving > shadows, difference in lighting, vibration, etc. > > >> then I have to send new picture out thru >> wireless to pc. what about PIC24EP? > > > They can do 60MIPS. Just to scan two 5MP images and compare them would > take 0.25s *IF* you could access each pixel in one cycle and do the > compare in one cycle. > As far as I know, no PIC or dsPIC have external bus interface that can > access such a large memory (not even they have SDRAM controllers), and > surely not at one transfer per clock. > If you implement your own bus, it will be much slower. My implementation > takes 5 instructions to do a transfer and it is optimized as hell, and > the address generation is done by the CPLD, The dsPIC just needs to > mange the data and control lines. > > > As I said, you need a large MPU (100's of MIPS) with some MB of SDRAM > and a large cache. > > > Best regards, > > Isaac > > >> >> thanks >> >> Andre >> >> >> >> >> ________________________________ >> From: Isaac Marino Bavaresco <@spam@isaacbavaresco@spam@spam_OUTyahoo.com.br> >> To: Microcontroller discussion list - Public. <spamBeGonepiclistKILLspammit.edu> >> Sent: Monday, May 21, 2012 3:14 PM >> Subject: Re: [PIC]: PIC32 for camera capture >> >> SUPPOSING that the PIC can DMA transfer data from the sensor to an >> external memory in one cycle (it can't), the maximum frame rate you >> could accomplish would be 16fps at 80MHz, and you would have no time to >> process it. >> Usually, image processing is much more intensive than the capture (that >> is, you will need to access each pixel several times). >> >> Your external memory needs to be at least 10MiB (16MiB is the logical >> choice) for color images (YCbCr) or 5MiB (8MiB being the logical choice) >> for gray scale images, just to hold one frame. >> >> Do you really need such a large image? We do fingerprint recognition >> with VGA sensors (640x480) cropped to 440x380. >> We use a dsPIC to pre-process the image but the capture is done with a >> Xilinx CPLD coupled to a 256KiBx8 static RAM. >> We can capture approx. 13 frames per second and can determine if a valid >> fingerprint is present in less than 1/2 second. >> >> >> Best regards, >> >> Isaac >> >> >> >> Em 21/5/2012 18:13, Andre Abelian escreveu: >>> Hi all, >>> >>> does any one know if pic32 will have enough speed to capture a camera data? >>> I am thinking to use 5 mp camera directly connected to pic and of course there is >>> a RAM too for a buffer. main purpose of the camera is to sense a motion change and >>> send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA >>> part for some how I believe PIC32 may do the job. once the motion is detected we can stop it >>> send it out then continue. I do not need to do continues live video transmit or some thing. >>> >>> any idea or suggestion will appreciate >>> >>> thanks >>> >>> Andre > >

You don't need dual port RAM. Implement a counter in the CPLD to generate the RAM address and increment it after each access. Provide some means for the PIC to reset the counter. Connect the pixel clock from the sensor to trigger the memory writes and increment the address counter. Probably that logic must include HSYNC and VSYNC also. Add a signal for the CPLD signal that a frame is finished and another to start and cancel the capture by the PIC. It would be a bonus if you can read and modify the address counter also, and even better if the PIC can also write to the RAM. Don't forget that your logic must wait for the start of a frame before it starts to write to the RAM, and it must stop when the frame is done. My design does all that and more, so I can use some of the RAM for application data. I use SRAM. As you need SDRAM, perhaps a CPLD is not enough. Probably you need an FPGA. Isaac Em 23/5/2012 12:24, Andre Abelian escreveu: {Quote hidden}> Isaac, > > I agree. My question is let say I use cpld to capture the data now i need to compare portion of the image to > determine as motion change. I am thinking to use dual ported ram to capture the image and use pic to access > to ram and compare data? what do you think? > > Andre > > > > ________________________________ > From: Isaac Marino Bavaresco <.....isaacbavarescospam_OUTyahoo.com.br> > To: Microcontroller discussion list - Public. <TakeThisOuTpiclist.....TakeThisOuTmit.edu> > Sent: Wednesday, May 23, 2012 5:56 AM > Subject: Re: [PIC]: PIC32 for camera capture > > Did you notice that 10ns is 100MHz? > > That means that you have to read one 12-bit pixel from the sensor and > store it to external memory faster than any PIC can execute one instruction. > Even a 400MHz ARM would have a hard time doing this without hardware > assistance. I tried and desisted. It is doable, but the ARM becomes > enslaved by the image capture. > > > Isaac > > > > > > Em 22/5/2012 23:33, Andre Abelian escreveu: >> Hi Jason, >> >> Finally I chose the part MT9P401 from Aptina. based on my understanding >> the pixel output period is every 10.4nS. so the pic should be able to capture 12 bit "12 pins" data ever >> 10 nS and save it in SDRAM. basically there are 5.000.000 pixels I need to capture and store them in SDRAM. >> how about pic24EP 120mhz is it faster then pic32? in my case I really need 16 bit. >> pic32 is 80mhz/2 40 mhz and pic24EP is 120mhz/2 60 mhz. what do you think? >> >> any suggestion will appreciate >> >> thanks >> >> Andre. >> >> >> ________________________________ >> From: Jason White <TakeThisOuTwhitewaterssoftwareinfoKILLspamspamgmail.com> >> To: Microcontroller discussion list - Public. <.....piclistRemoveMEmit.edu> >> Sent: Monday, May 21, 2012 2:25 PM >> Subject: Re: [PIC]: PIC32 for camera capture >> >> On Mon, May 21, 2012 at 5:13 PM, Andre Abelian <RemoveMEabelian.andrespamBeGoneyahoo.com> wrote: >>> Hi all, >>> >>> does any one know if pic32 will have enough speed to capture a camera data? >>> I am thinking to use 5 mp camera directly connected to pic and of course there is >>> a RAM too for a buffer. main purpose of the camera is to sense a motion change and >>> send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA >>> part for some how I believe PIC32 may do the job. once the motion is detected we can stop it >>> send it out then continue. I do not need to do continues live video transmit or some thing. >>> >>> any idea or suggestion will appreciate >> I am no expert at this, but the frame rate you want to read from the >> camera and process the data will determine whether or not your >> processor is fast enough. More details please ! >> >>

Isaac, You mentioned 10mb for grayscale and 20 mb for color. is't single reading of each pixel has all information? if I scan twice image may change. my understanding was single scan of each pixel will give all information but I am not sure this is my first project to learn. what do you think? AA ________________________________ From: Isaac Marino Bavaresco <spamBeGoneisaacbavaresco@spam@spam_OUTyahoo.com.br> To: Microcontroller discussion list - Public. <TakeThisOuTpiclistspammit.edu> Sent: Wednesday, May 23, 2012 3:48 AM Subject: Re: [PIC]: PIC32 for camera capture Em 22/5/2012 23:47, Andre Abelian escreveu: > Hi Isaac, > > I didn't see your replay for some reason. the purpose of this camera is to capture > a person and send email as attachment. the quality is important. in your case how pic is > communicating with CPLD? I implemented a bit-banged 8-bit bus. It can transfer 8MB/s. > in my case the only part I am not clear is I have to sense a motion. You need to store at least two images (10MB for B/W images, 20MB for color images) and compare them. The biggest problem I see is that you cannot simply bytewise compare the images, you must use some intelligent algorithm to detect real movement, not noise or small changes in lighting. Prepare to scan both images several times (read each pixel more than once) and do a lot of math (FFT, filtering, wavelet, etc.). I think that this task is up to a much larger processor (ARM9 @ 400MIPS, etc). > if old picture and new are not same They will never be the same, always there are a lot of noise, moving shadows, difference in lighting, vibration, etc. > then I have to send new picture out thru > wireless to pc. what about PIC24EP? They can do 60MIPS. Just to scan two 5MP images and compare them would take 0.25s *IF* you could access each pixel in one cycle and do the compare in one cycle. As far as I know, no PIC or dsPIC have external bus interface that can access such a large memory (not even they have SDRAM controllers), and surely not at one transfer per clock. If you implement your own bus, it will be much slower. My implementation takes 5 instructions to do a transfer and it is optimized as hell, and the address generation is done by the CPLD, The dsPIC just needs to mange the data and control lines. As I said, you need a large MPU (100's of MIPS) with some MB of SDRAM and a large cache. Best regards, Isaac {Quote hidden}> > thanks > > Andre > > > > > ________________________________ > From: Isaac Marino Bavaresco <isaacbavarescoEraseMEyahoo.com.br> > To: Microcontroller discussion list - Public. <RemoveMEpiclistEraseMEspam_OUTmit.edu> > Sent: Monday, May 21, 2012 3:14 PM > Subject: Re: [PIC]: PIC32 for camera capture > > SUPPOSING that the PIC can DMA transfer data from the sensor to an > external memory in one cycle (it can't), the maximum frame rate you > could accomplish would be 16fps at 80MHz, and you would have no time to > process it. > Usually, image processing is much more intensive than the capture (that > is, you will need to access each pixel several times). > > Your external memory needs to be at least 10MiB (16MiB is the logical > choice) for color images (YCbCr) or 5MiB (8MiB being the logical choice) > for gray scale images, just to hold one frame. > > Do you really need such a large image? We do fingerprint recognition > with VGA sensors (640x480) cropped to 440x380. > We use a dsPIC to pre-process the image but the capture is done with a > Xilinx CPLD coupled to a 256KiBx8 static RAM. > We can capture approx. 13 frames per second and can determine if a valid > fingerprint is present in less than 1/2 second. > > > Best regards, > > Isaac > > > > Em 21/5/2012 18:13, Andre Abelian escreveu: >> Hi all, >> >> does any one know if pic32 will have enough speed to capture a camera data? >> I am thinking to use 5 mp camera directly connected to pic and of course there is >> a RAM too for a buffer. main purpose of the camera is to sense a motion change and >> send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA >> part for some how I believe PIC32 may do the job. once the motion is detected we can stop it >> send it out then continue. I do not need to do continues live video transmit or some thing. >> >> any idea or suggestion will appreciate >> >> thanks >> >> Andre

As Isaac stated, you don't need dual ram, it all can be done in CPLD. 5MP sounds big to me, why do you think you need 5MP? What kind of algorithm are you planning to use for comparison ? On Wed, May 23, 2012 at 6:30 PM, Andre Abelian <@spam@abelian.andreRemoveMEEraseMEyahoo.com> wrote: {Quote hidden}> Yigit, > > the way I am seeing iti need to use cpld to capture the image in background and > use pic to access to ram using dual ported ram that way motion can be detected > like you suggested using gray scale portion. Now my question is for pic do I need > ram? or I can use internal program memory instead? > > thanks > > AA > > > > > ________________________________ > From: Yigit Turgut <EraseMEy.turgut@spam@gmail.com> > To: Microcontroller discussion list - Public. <@spam@piclistspam_OUT.....mit.edu> > Sent: Wednesday, May 23, 2012 4:25 AM > Subject: Re: [PIC]: PIC32 for camera capture > > I don't think PIC is your way to go in this case. You might be able to > detect the motion in grayscale precisely, but face recognition, person > identification from image etc can not be done real-time via PIC. As > others stated you can check out ARM9-10-11. Why don't you switch to > FPGA ? An entry level FPGA will be capable of doing these kinds of > tasks very well without that much of an effort. > > On Wed, May 23, 2012 at 1:48 PM, Isaac Marino Bavaresco > <spamBeGoneisaacbavarescoEraseMEyahoo.com.br> wrote: >> Em 22/5/2012 23:47, Andre Abelian escreveu: >>> Hi Isaac, >>> >>> I didn't see your replay for some reason. the purpose of this camera is to capture >>> a person and send email as attachment. the quality is important. in your case how pic is >>> communicating with CPLD? >> >> I implemented a bit-banged 8-bit bus. It can transfer 8MB/s. >> >> >>> in my case the only part I am not clear is I have to sense a motion.. >> >> >> You need to store at least two images (10MB for B/W images, 20MB for >> color images) and compare them. >> The biggest problem I see is that you cannot simply bytewise compare the >> images, you must use some intelligent algorithm to detect real movement, >> not noise or small changes in lighting. >> Prepare to scan both images several times (read each pixel more than >> once) and do a lot of math (FFT, filtering, wavelet, etc.). >> I think that this task is up to a much larger processor (ARM9 @ 400MIPS, >> etc). >> >> >>> if old picture and new are not same >> >> >> They will never be the same, always there are a lot of noise, moving >> shadows, difference in lighting, vibration, etc. >> >> >>> then I have to send new picture out thru >>> wireless to pc. what about PIC24EP? >> >> >> They can do 60MIPS. Just to scan two 5MP images and compare them would >> take 0.25s *IF* you could access each pixel in one cycle and do the >> compare in one cycle. >> As far as I know, no PIC or dsPIC have external bus interface that can >> access such a large memory (not even they have SDRAM controllers), and >> surely not at one transfer per clock. >> If you implement your own bus, it will be much slower. My implementation >> takes 5 instructions to do a transfer and it is optimized as hell, and >> the address generation is done by the CPLD, The dsPIC just needs to >> mange the data and control lines. >> >> >> As I said, you need a large MPU (100's of MIPS) with some MB of SDRAM >> and a large cache. >> >> >> Best regards, >> >> Isaac >> >> >>> >>> thanks >>> >>> Andre >>> >>> >>> >>> >>> ________________________________ >>> From: Isaac Marino Bavaresco <isaacbavarescospamBeGoneyahoo.com.br> >>> To: Microcontroller discussion list - Public. <RemoveMEpiclist@spam@spamBeGonemit.edu> >>> Sent: Monday, May 21, 2012 3:14 PM >>> Subject: Re: [PIC]: PIC32 for camera capture >>> >>> SUPPOSING that the PIC can DMA transfer data from the sensor to an >>> external memory in one cycle (it can't), the maximum frame rate you >>> could accomplish would be 16fps at 80MHz, and you would have no time to >>> process it. >>> Usually, image processing is much more intensive than the capture (that >>> is, you will need to access each pixel several times). >>> >>> Your external memory needs to be at least 10MiB (16MiB is the logical >>> choice) for color images (YCbCr) or 5MiB (8MiB being the logical choice) >>> for gray scale images, just to hold one frame. >>> >>> Do you really need such a large image? We do fingerprint recognition >>> with VGA sensors (640x480) cropped to 440x380. >>> We use a dsPIC to pre-process the image but the capture is done with a >>> Xilinx CPLD coupled to a 256KiBx8 static RAM. >>> We can capture approx. 13 frames per second and can determine if a valid >>> fingerprint is present in less than 1/2 second. >>> >>> >>> Best regards, >>> >>> Isaac >>> >>> >>> >>> Em 21/5/2012 18:13, Andre Abelian escreveu: >>>> Hi all, >>>> >>>> does any one know if pic32 will have enough speed to capture a camera data? >>>> I am thinking to use 5 mp camera directly connected to pic and of course there is >>>> a RAM too for a buffer. main purpose of the camera is to sense a motion change and >>>> send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA >>>> part for some how I believe PIC32 may do the job. once the motion is detected we can stop it >>>> send it out then continue. I do not need to do continues live video transmit or some thing. >>>> >>>> any idea or suggestion will appreciate >>>> >>>> thanks >>>> >>>> Andre >> >> -

Andre, It seems that you are a little lost with this design. The PIC with largest RAM have 128kB, you need at least 10MB. And there is no way for a CPLD or whatever push data at 100MB/s into the PIC. You need to capture to an external memory using an autonomous frame grabber and then process the images. Understand that the image processing will take orders of magnitude more time than the capture process. Isaac Em 23/5/2012 12:30, Andre Abelian escreveu: {Quote hidden}> Yigit, > > the way I am seeing iti need to use cpld to capture the image in background and > use pic to access to ram using dual ported ram that way motion can be detected > like you suggested using gray scale portion. Now my question is for pic do I need > ram? or I can use internal program memory instead? > > thanks > > AA > > > > > ________________________________ > From: Yigit Turgut <y.turgutEraseME@spam@gmail.com> > To: Microcontroller discussion list - Public. <RemoveMEpiclistspamBeGonemit.edu> > Sent: Wednesday, May 23, 2012 4:25 AM > Subject: Re: [PIC]: PIC32 for camera capture > > I don't think PIC is your way to go in this case. You might be able to > detect the motion in grayscale precisely, but face recognition, person > identification from image etc can not be done real-time via PIC. As > others stated you can check out ARM9-10-11. Why don't you switch to > FPGA ? An entry level FPGA will be capable of doing these kinds of > tasks very well without that much of an effort. > > On Wed, May 23, 2012 at 1:48 PM, Isaac Marino Bavaresco > <spamBeGoneisaacbavarescoKILLspam@spam@yahoo.com.br> wrote: >> Em 22/5/2012 23:47, Andre Abelian escreveu: >>> Hi Isaac, >>> >>> I didn't see your replay for some reason. the purpose of this camera is to capture >>> a person and send email as attachment. the quality is important. in your case how pic is >>> communicating with CPLD? >> I implemented a bit-banged 8-bit bus. It can transfer 8MB/s. >> >> >>> in my case the only part I am not clear is I have to sense a motion. >> >> You need to store at least two images (10MB for B/W images, 20MB for >> color images) and compare them. >> The biggest problem I see is that you cannot simply bytewise compare the >> images, you must use some intelligent algorithm to detect real movement, >> not noise or small changes in lighting. >> Prepare to scan both images several times (read each pixel more than >> once) and do a lot of math (FFT, filtering, wavelet, etc.). >> I think that this task is up to a much larger processor (ARM9 @ 400MIPS, >> etc). >> >> >>> if old picture and new are not same >> >> They will never be the same, always there are a lot of noise, moving >> shadows, difference in lighting, vibration, etc. >> >> >>> then I have to send new picture out thru >>> wireless to pc. what about PIC24EP? >> >> They can do 60MIPS. Just to scan two 5MP images and compare them would >> take 0.25s *IF* you could access each pixel in one cycle and do the >> compare in one cycle. >> As far as I know, no PIC or dsPIC have external bus interface that can >> access such a large memory (not even they have SDRAM controllers), and >> surely not at one transfer per clock. >> If you implement your own bus, it will be much slower. My implementation >> takes 5 instructions to do a transfer and it is optimized as hell, and >> the address generation is done by the CPLD, The dsPIC just needs to >> mange the data and control lines. >> >> >> As I said, you need a large MPU (100's of MIPS) with some MB of SDRAM >> and a large cache. >> >> >> Best regards, >> >> Isaac >> >> >>> thanks >>> >>> Andre >>> >>> >>> >>> >>> ________________________________ >>> From: Isaac Marino Bavaresco <isaacbavarescospam_OUT@spam@yahoo.com.br> >>> To: Microcontroller discussion list - Public. <spamBeGonepiclist@spam@mit.edu> >>> Sent: Monday, May 21, 2012 3:14 PM >>> Subject: Re: [PIC]: PIC32 for camera capture >>> >>> SUPPOSING that the PIC can DMA transfer data from the sensor to an >>> external memory in one cycle (it can't), the maximum frame rate you >>> could accomplish would be 16fps at 80MHz, and you would have no time to >>> process it. >>> Usually, image processing is much more intensive than the capture (that >>> is, you will need to access each pixel several times). >>> >>> Your external memory needs to be at least 10MiB (16MiB is the logical >>> choice) for color images (YCbCr) or 5MiB (8MiB being the logical choice) >>> for gray scale images, just to hold one frame. >>> >>> Do you really need such a large image? We do fingerprint recognition >>> with VGA sensors (640x480) cropped to 440x380. >>> We use a dsPIC to pre-process the image but the capture is done with a >>> Xilinx CPLD coupled to a 256KiBx8 static RAM. >>> We can capture approx. 13 frames per second and can determine if a valid >>> fingerprint is present in less than 1/2 second. >>> >>> >>> Best regards, >>> >>> Isaac >>> >>> >>> >>> Em 21/5/2012 18:13, Andre Abelian escreveu: >>>> Hi all, >>>> >>>> does any one know if pic32 will have enough speed to capture a camera data? >>>> I am thinking to use 5 mp camera directly connected to pic and of course there is >>>> a RAM too for a buffer. main purpose of the camera is to sense a motion change and >>>> send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA >>>> part for some how I believe PIC32 may do the job. once the motion is detected we can stop it >>>> send it out then continue. I do not need to do continues live video transmit or some thing. >>>> >>>> any idea or suggestion will appreciate >>>> >>>> thanks >>>> >>>> Andre >> -

You want to compare two images, so you must hold both in RAM. They may be two consecutive images or a base image plus another taken some time in the future. Each image has 5,000,000 pixels. A grayscale image uses one byte per pixel, a color image uses two bytes per pixel. So a grayscale image needs 5,000,000 bytes of RAM and a color image needs 10,000,000 bytes of RAM. Thus the figures, 10MB for grayscale images and 20MB for color images. For capturing, each pixel is read once and stored, but the comparison algorithm may need to do several passes over the captured data. Isaac Em 23/5/2012 13:19, Andre Abelian escreveu: {Quote hidden}> Isaac, > > You mentioned 10mb for grayscale and 20 mb for color. is't single reading of > each pixel has all information? if I scan twice image may change. my understanding was > single scan of each pixel will give all information but I am not sure this is my first project > to learn. > > what do you think? > > AA > > > ________________________________ > From: Isaac Marino Bavaresco <RemoveMEisaacbavarescoEraseMEKILLspamyahoo.com.br> > To: Microcontroller discussion list - Public. <spamBeGonepiclistspam_OUTRemoveMEmit.edu> > Sent: Wednesday, May 23, 2012 3:48 AM > Subject: Re: [PIC]: PIC32 for camera capture > > Em 22/5/2012 23:47, Andre Abelian escreveu: >> Hi Isaac, >> >> I didn't see your replay for some reason. the purpose of this camera is to capture >> a person and send email as attachment. the quality is important. in your case how pic is >> communicating with CPLD? > I implemented a bit-banged 8-bit bus. It can transfer 8MB/s. > > >> in my case the only part I am not clear is I have to sense a motion. > > You need to store at least two images (10MB for B/W images, 20MB for > color images) and compare them. > The biggest problem I see is that you cannot simply bytewise compare the > images, you must use some intelligent algorithm to detect real movement, > not noise or small changes in lighting. > Prepare to scan both images several times (read each pixel more than > once) and do a lot of math (FFT, filtering, wavelet, etc.). > I think that this task is up to a much larger processor (ARM9 @ 400MIPS, > etc). > > >> if old picture and new are not same > > They will never be the same, always there are a lot of noise, moving > shadows, difference in lighting, vibration, etc. > > >> then I have to send new picture out thru >> wireless to pc. what about PIC24EP? > > They can do 60MIPS. Just to scan two 5MP images and compare them would > take 0.25s *IF* you could access each pixel in one cycle and do the > compare in one cycle. > As far as I know, no PIC or dsPIC have external bus interface that can > access such a large memory (not even they have SDRAM controllers), and > surely not at one transfer per clock. > If you implement your own bus, it will be much slower. My implementation > takes 5 instructions to do a transfer and it is optimized as hell, and > the address generation is done by the CPLD, The dsPIC just needs to > mange the data and control lines. > > > As I said, you need a large MPU (100's of MIPS) with some MB of SDRAM > and a large cache. > > > Best regards, > > Isaac > > >> thanks >> >> Andre >> >> >> >> >> ________________________________ >> From: Isaac Marino Bavaresco <.....isaacbavarescoRemoveMEyahoo.com.br> >> To: Microcontroller discussion list - Public. <piclist@spam@mit.edu> >> Sent: Monday, May 21, 2012 3:14 PM >> Subject: Re: [PIC]: PIC32 for camera capture >> >> SUPPOSING that the PIC can DMA transfer data from the sensor to an >> external memory in one cycle (it can't), the maximum frame rate you >> could accomplish would be 16fps at 80MHz, and you would have no time to >> process it. >> Usually, image processing is much more intensive than the capture (that >> is, you will need to access each pixel several times). >> >> Your external memory needs to be at least 10MiB (16MiB is the logical >> choice) for color images (YCbCr) or 5MiB (8MiB being the logical choice) >> for gray scale images, just to hold one frame. >> >> Do you really need such a large image? We do fingerprint recognition >> with VGA sensors (640x480) cropped to 440x380. >> We use a dsPIC to pre-process the image but the capture is done with a >> Xilinx CPLD coupled to a 256KiBx8 static RAM. >> We can capture approx. 13 frames per second and can determine if a valid >> fingerprint is present in less than 1/2 second. >> >> >> Best regards, >> >> Isaac >> >> >> >> Em 21/5/2012 18:13, Andre Abelian escreveu: >>> Hi all, >>> >>> does any one know if pic32 will have enough speed to capture a camera data? >>> I am thinking to use 5 mp camera directly connected to pic and of course there is >>> a RAM too for a buffer. main purpose of the camera is to sense a motion change and >>> send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA >>> part for some how I believe PIC32 may do the job. once the motion is detected we can stop it >>> send it out then continue. I do not need to do continues live video transmit or some thing. >>> >>> any idea or suggestion will appreciate >>> >>> thanks >>> >>> Andre

Isaac, I guess I didn't explain clearly. the purpose of pic is to compare 2 pictures but not full picture. lets say small area in the center half inch square. I am almost thinking that separate PIR sensor is better idea. image comparing is workable but needs better understanding and experiment. The way I am seeing it there are lots of situations where it may create false triggering on e of them is darkness may add noise and create false triggering. Isaac, what I do not get it is that in datasheet only says each pixel is 12 bit wide and I do not see how to decode? you are saying I have to read a few times in order to get gray scale and color. based on datasheet pixel bit will output every 10.4 ns and on falling edge read 12 pins output data. you said 1 byte for grayscale 2 bytes for color that 24 bit data. my question is how to get 24 bit data if it outputs only 12bit. thanks Andre ________________________________ From: Isaac Marino Bavaresco <EraseMEisaacbavarescoRemoveMESTOPspamyahoo.com.br> To: Microcontroller discussion list - Public. <RemoveMEpiclistKILLspamTakeThisOuTmit.edu> Sent: Wednesday, May 23, 2012 9:30 AM Subject: Re: [PIC]: PIC32 for camera capture Andre, It seems that you are a little lost with this design. The PIC with largest RAM have 128kB, you need at least 10MB. And there is no way for a CPLD or whatever push data at 100MB/s into the PIC. You need to capture to an external memory using an autonomous frame grabber and then process the images. Understand that the image processing will take orders of magnitude more time than the capture process. Isaac Em 23/5/2012 12:30, Andre Abelian escreveu: {Quote hidden}> Yigit, > > the way I am seeing iti need to use cpld to capture the image in background and > use pic to access to ram using dual ported ram that way motion can be detected > like you suggested using gray scale portion. Now my question is for pic do I need > ram? or I can use internal program memory instead? > > thanks > > AA > > > > > ________________________________ > From: Yigit Turgut <spamBeGoney.turgut@spam@gmail.com> > To: Microcontroller discussion list - Public. <RemoveMEpiclistspam_OUTmit.edu> > Sent: Wednesday, May 23, 2012 4:25 AM > Subject: Re: [PIC]: PIC32 for camera capture > > I don't think PIC is your way to go in this case. You might be able to > detect the motion in grayscale precisely, but face recognition, person > identification from image etc can not be done real-time via PIC. As > others stated you can check out ARM9-10-11. Why don't you switch to > FPGA ? An entry level FPGA will be capable of doing these kinds of > tasks very well without that much of an effort. > > On Wed, May 23, 2012 at 1:48 PM, Isaac Marino Bavaresco > <isaacbavarescospamyahoo.com.br> wrote: >> Em 22/5/2012 23:47, Andre Abelian escreveu: >>> Hi Isaac, >>> >>> I didn't see your replay for some reason. the purpose of this camera is to capture >>> a person and send email as attachment. the quality is important. in your case how pic is >>> communicating with CPLD? >> I implemented a bit-banged 8-bit bus. It can transfer 8MB/s. >> >> >>> in my case the only part I am not clear is I have to sense a motion. >> >> You need to store at least two images (10MB for B/W images, 20MB for >> color images) and compare them. >> The biggest problem I see is that you cannot simply bytewise compare the >> images, you must use some intelligent algorithm to detect real movement, >> not noise or small changes in lighting. >> Prepare to scan both images several times (read each pixel more than >> once) and do a lot of math (FFT, filtering, wavelet, etc.). >> I think that this task is up to a much larger processor (ARM9 @ 400MIPS, >> etc). >> >> >>> if old picture and new are not same >> >> They will never be the same, always there are a lot of noise, moving >> shadows, difference in lighting, vibration, etc. >> >> >>> then I have to send new picture out thru >>> wireless to pc. what about PIC24EP? >> >> They can do 60MIPS. Just to scan two 5MP images and compare them would >> take 0.25s *IF* you could access each pixel in one cycle and do the >> compare in one cycle. >> As far as I know, no PIC or dsPIC have external bus interface that can >> access such a large memory (not even they have SDRAM controllers), and >> surely not at one transfer per clock. >> If you implement your own bus, it will be much slower. My implementation >> takes 5 instructions to do a transfer and it is optimized as hell, and >> the address generation is done by the CPLD, The dsPIC just needs to >> mange the data and control lines. >> >> >> As I said, you need a large MPU (100's of MIPS) with some MB of SDRAM >> and a large cache. >> >> >> Best regards, >> >> Isaac >> >> >>> thanks >>> >>> Andre >>> >>> >>> >>> >>> ________________________________ >>> From: Isaac Marino Bavaresco <spam_OUTisaacbavarescospam_OUTspam_OUTyahoo.com.br> >>> To: Microcontroller discussion list - Public. <piclistspam_OUTmit.edu> >>> Sent: Monday, May 21, 2012 3:14 PM >>> Subject: Re: [PIC]: PIC32 for camera capture >>> >>> SUPPOSING that the PIC can DMA transfer data from the sensor to an >>> external memory in one cycle (it can't), the maximum frame rate you >>> could accomplish would be 16fps at 80MHz, and you would have no time to >>> process it. >>> Usually, image processing is much more intensive than the capture (that >>> is, you will need to access each pixel several times). >>> >>> Your external memory needs to be at least 10MiB (16MiB is the logical >>> choice) for color images (YCbCr) or 5MiB (8MiB being the logical choice) >>> for gray scale images, just to hold one frame. >>> >>> Do you really need such a large image? We do fingerprint recognition >>> with VGA sensors (640x480) cropped to 440x380. >>> We use a dsPIC to pre-process the image but the capture is done with a >>> Xilinx CPLD coupled to a 256KiBx8 static RAM. >>> We can capture approx. 13 frames per second and can determine if a valid >>> fingerprint is present in less than 1/2 second. >>> >>> >>> Best regards, >>> >>> Isaac >>> >>> >>> >>> Em 21/5/2012 18:13, Andre Abelian escreveu: >>>> Hi all, >>>> >>>> does any one know if pic32 will have enough speed to capture a camera data? >>>> I am thinking to use 5 mp camera directly connected to pic and of course there is >>>> a RAM too for a buffer. main purpose of the camera is to sense a motion change and >>>> send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA >>>> part for some how I believe PIC32 may do the job. once the motion is detected we can stop it >>>> send it out then continue. I do not need to do continues live video transmit or some thing. >>>> >>>> any idea or suggestion will appreciate >>>> >>>> thanks >>>> >>>> Andre >> -

Em 23/5/2012 16:32, Andre Abelian escreveu: > Isaac, > > I guess I didn't explain clearly. the purpose of pic is to compare 2 pictures but > not full picture. lets say small area in the center half inch square. I am almost thinking that separate > PIR sensor is better idea. image comparing is workable but needs better understanding and experiment. > The way I am seeing it there are lots of situations where it may create false triggering on e of them is > > darkness may add noise and create false triggering. > > Isaac, what I do not get it is that in datasheet only says each pixel is 12 bit wide and I do not > see how to decode? you are saying I have to read a few times in order to get gray scale and color. > based on datasheet pixel bit will output every 10.4 ns and on falling edge read 12 pins output data. > you said 1 byte for grayscale 2 bytes for color that 24 bit data. my question is how to get 24 bit data > if it outputs only 12bit. Your sensor is 12-bit, so each "byte" will be 12-bit. When I say "grayscale" image, I'm talking about a B/W sensor (or a color one that can be configured to output B/W images) that outputs only the luminance (Y) component of image. The image is composed of a sequence of 12-bit values representing the brightness of each pixel. A color sensor outputs the image as a sequence of two values (12-bit each in your case), first the luminance (Y) and then the color component (CbCr) for each pixel. Some color sensors can output the image in RGB format (usually 5:6:5 bits) or in the YUV format. In a color sensor in YCbCr or YUV mode, if you discard the second value, you get the grayscale image. Isaac {Quote hidden}> > thanks > > Andre > > > > > > > > > ________________________________ > From: Isaac Marino Bavaresco <RemoveMEisaacbavarescoKILLspam@spam@yahoo.com.br> > To: Microcontroller discussion list - Public. <piclistspamBeGone.....mit.edu> > Sent: Wednesday, May 23, 2012 9:30 AM > Subject: Re: [PIC]: PIC32 for camera capture > > Andre, > > > It seems that you are a little lost with this design. The PIC with > largest RAM have 128kB, you need at least 10MB. > And there is no way for a CPLD or whatever push data at 100MB/s into the > PIC. > > You need to capture to an external memory using an autonomous frame > grabber and then process the images. > > Understand that the image processing will take orders of magnitude more > time than the capture process. > > > Isaac > > > > Em 23/5/2012 12:30, Andre Abelian escreveu: >> Yigit, >> >> the way I am seeing iti need to use cpld to capture the image in background and >> use pic to access to ram using dual ported ram that way motion can be detected >> like you suggested using gray scale portion. Now my question is for pic do I need >> ram? or I can use internal program memory instead? >> >> thanks >> >> AA >> >> >> >> >> ________________________________ >> From: Yigit Turgut <KILLspamy.turgut.....gmail.com> >> To: Microcontroller discussion list - Public. <spam_OUTpiclistKILLspammit.edu> >> Sent: Wednesday, May 23, 2012 4:25 AM >> Subject: Re: [PIC]: PIC32 for camera capture >> >> I don't think PIC is your way to go in this case. You might be able to >> detect the motion in grayscale precisely, but face recognition, person >> identification from image etc can not be done real-time via PIC. As >> others stated you can check out ARM9-10-11. Why don't you switch to >> FPGA ? An entry level FPGA will be capable of doing these kinds of >> tasks very well without that much of an effort. >> >> On Wed, May 23, 2012 at 1:48 PM, Isaac Marino Bavaresco >> <RemoveMEisaacbavarescoRemoveMEEraseMEyahoo.com.br> wrote: >>> Em 22/5/2012 23:47, Andre Abelian escreveu: >>>> Hi Isaac, >>>> >>>> I didn't see your replay for some reason. the purpose of this camera is to capture >>>> a person and send email as attachment. the quality is important. in your case how pic is >>>> communicating with CPLD? >>> I implemented a bit-banged 8-bit bus. It can transfer 8MB/s. >>> >>> >>>> in my case the only part I am not clear is I have to sense a motion.. >>> You need to store at least two images (10MB for B/W images, 20MB for >>> color images) and compare them. >>> The biggest problem I see is that you cannot simply bytewise compare the >>> images, you must use some intelligent algorithm to detect real movement, >>> not noise or small changes in lighting. >>> Prepare to scan both images several times (read each pixel more than >>> once) and do a lot of math (FFT, filtering, wavelet, etc.). >>> I think that this task is up to a much larger processor (ARM9 @ 400MIPS, >>> etc). >>> >>> >>>> if old picture and new are not same >>> They will never be the same, always there are a lot of noise, moving >>> shadows, difference in lighting, vibration, etc. >>> >>> >>>> then I have to send new picture out thru >>>> wireless to pc. what about PIC24EP? >>> They can do 60MIPS. Just to scan two 5MP images and compare them would >>> take 0.25s *IF* you could access each pixel in one cycle and do the >>> compare in one cycle. >>> As far as I know, no PIC or dsPIC have external bus interface that can >>> access such a large memory (not even they have SDRAM controllers), and >>> surely not at one transfer per clock. >>> If you implement your own bus, it will be much slower. My implementation >>> takes 5 instructions to do a transfer and it is optimized as hell, and >>> the address generation is done by the CPLD, The dsPIC just needs to >>> mange the data and control lines. >>> >>> >>> As I said, you need a large MPU (100's of MIPS) with some MB of SDRAM >>> and a large cache. >>> >>> >>> Best regards, >>> >>> Isaac >>> >>> >>>> thanks >>>> >>>> Andre >>>> >>>> >>>> >>>> >>>> ________________________________ >>>> From: Isaac Marino Bavaresco <KILLspamisaacbavarescospamBeGoneyahoo.com.br> >>>> To: Microcontroller discussion list - Public. <piclistspammit.edu> >>>> Sent: Monday, May 21, 2012 3:14 PM >>>> Subject: Re: [PIC]: PIC32 for camera capture >>>> >>>> SUPPOSING that the PIC can DMA transfer data from the sensor to an >>>> external memory in one cycle (it can't), the maximum frame rate you >>>> could accomplish would be 16fps at 80MHz, and you would have no time to >>>> process it. >>>> Usually, image processing is much more intensive than the capture (that >>>> is, you will need to access each pixel several times). >>>> >>>> Your external memory needs to be at least 10MiB (16MiB is the logical >>>> choice) for color images (YCbCr) or 5MiB (8MiB being the logical choice) >>>> for gray scale images, just to hold one frame. >>>> >>>> Do you really need such a large image? We do fingerprint recognition >>>> with VGA sensors (640x480) cropped to 440x380. >>>> We use a dsPIC to pre-process the image but the capture is done with a >>>> Xilinx CPLD coupled to a 256KiBx8 static RAM. >>>> We can capture approx. 13 frames per second and can determine if a valid >>>> fingerprint is present in less than 1/2 second. >>>> >>>> >>>> Best regards, >>>> >>>> Isaac >>>> >>>> >>>> >>>> Em 21/5/2012 18:13, Andre Abelian escreveu: >>>>> Hi all, >>>>> >>>>> does any one know if pic32 will have enough speed to capture a camera data? >>>>> I am thinking to use 5 mp camera directly connected to pic and of course there is >>>>> a RAM too for a buffer. main purpose of the camera is to sense a motion change and >>>>> send the data thru wireless when change detected. I am originally picked altera cyclone3 FPGA >>>>> part for some how I believe PIC32 may do the job. once the motion is detected we can stop it >>>>> send it out then continue. I do not need to do continues live video transmit or some thing. >>>>> >>>>> any idea or suggestion will appreciate >>>>> >>>>> thanks >>>>> >>>>> Andre >>> --