SI-1920HD and strobe flash at DVinfo.net

Martin Breidt · May 23rd, 2007, 10:23 AM

I am thinking of integrating an SI-1920HD camera into our computer-controlled recording setup but I am unsure about how the rolling shutter of the camera will handle strobe flash illumination.

We are planning on running the camera at 60 frames/sec and will have an external trigger signal for the camera and the strobe, so the strobe will be in sync with the camera.

Having read about the camera's rolling shutter and how that means that each sensor line will start integrating at a slightly different point of time, I wonder whether there will be any point in time at which *all* parts of the sensor will collect light, at which the flash should be then triggered. Otherwise, I might end up with parts of the image properly exposed due to the flash, and parts of the image dark as the flash is already gone or has not yet fired.

Any insights much appreciated!
Thanks in advance
Martin

Jason Rodriguez · May 23rd, 2007, 01:07 PM

This can be done, but we would have to activate a special integration mode which would trigger the start of integration while the strobe is off, and not trigger the strobe until the entire frame is integrating, and then read-out the rest of the frame in the dark. I'm thinking though this would extend the frame-time beyond 1/60th of a second . . . but the chip can actually clock a bit higher, so depending on the length of exposure you need, I don't see this being impossible. Since it is a custom application though that would require custom firmware and software to be written, it would be a dedicated development effort, so you should probably contact Steve Nordhauser at (518) 279-9098 and discuss your needs with him to see what can realistically be done.

Martin Breidt · May 25th, 2007, 05:21 AM

Quote:

Originally Posted by Jason Rodriguez

This can be done, but we would have to activate a special integration mode which would trigger the start of integration while the strobe is off, and not trigger the strobe until the entire frame is integrating, and then read-out the rest of the frame in the dark. I'm thinking though this would extend the frame-time beyond 1/60th of a second . . . but the chip can actually clock a bit higher, so depending on the length of exposure you need, I don't see this being impossible. Since it is a custom application though that would require custom firmware and software to be written, it would be a dedicated development effort, so you should probably contact Steve Nordhauser at (518) 279-9098 and discuss your needs with him to see what can realistically be done.

Thanks Jason! I take it you are with SI?

Why does the length of exposure matter if the strobe is basically the only light available? I am still a little fuzzy on the rolling shutter.
What we could do on our side is solder up some triggering electronics that emit two seperate triggering impulses for camera and strobe.

Thanks again
Martin

Jason Rodriguez · May 25th, 2007, 11:44 PM

Basically the rolling shutter is very similiar to a focal plane array shutter on a still film camera, except that the rolling shutter on a CMOS moves from top->bottom rather than horizontally like on a film camera.

So if your strobe goes off when integration happens, then you will end up with what you mentioned before . . . part of the frame has integrated, and part of it hasn't. The ProCamHD, since it's a video-based chip, doesn't have a global reset release, meaning we can't start integration on every line at once. If it did, then the strobe and integration can fire off at the same time, since the entire chip would get exposed at once, and then the rolling shutter would do the read-out after the strobe has fired. Since it is a rolling shutter for both reset and read-out, you have to extend the horizontal blanking, so basically the chip resets every line and integration is happening simultaneously across the entire chip, you fire the strobe, and then the rolling shutter does the read-out. At least this is the way I understand things to happen. There may be some other tricks that can be done in firmware that I don't know about, but again, this would involve custom development, and Steve Nordhauser would be the best one to talk to about that at (518) 279-9098.

Ari Presler · May 26th, 2007, 01:05 PM

Martin,

We can simply provide you a strobe output signal from our cameras every top of frame. Set the exposure to full frame time (1/60-sec), which means all rows are integrating light, you will then have correctly exposed images.

Martin Breidt · May 30th, 2007, 02:41 AM

Quote:

Originally Posted by Ari Presler

Martin,

We can simply provide you a strobe output signal from our cameras every top of frame. Set the exposure to full frame time (1/60-sec), which means all rows are integrating light, you will then have correctly exposed images.

Ari,

thanks very much for the info. I am real sorry to be a pest but is there any chance you could sketch a diagram on how this works out temporally?
Because, as I understand it, even when each line is integrating, that does not necessarily mean there is one single point in time at which all lines are receiving light, right?

And if all lines are integrating for 1/60th of a second, there is time needed after integration for read-out and reset, correct? How much time is that minimum?

Not at all trying to say you do not know your own product :-), just trying to understand how the overall system design would look like with such a rolling shutter.

Maybe you could modify this image (http://www.siliconimaging.com/Images/SI-19228.gif) with a vertical line that indicates when the flash will be triggered?

Thanks again very much indeed!
Martin

Martin Breidt · May 30th, 2007, 02:49 AM

Quote:

Originally Posted by Jason Rodriguez

Basically the rolling shutter is very similiar to a focal plane array shutter on a still film camera, except that the rolling shutter on a CMOS moves from top->bottom rather than horizontally like on a film camera.

Fine, but a traditional camera has indeed problems with flash sync if the exposure is too short, because the first curtain is still moving when the second curtain is starting to move, so there is no point in time during the exposure that exposes all parts of the negative to light and therefore there is no point in time when a flash will evenly illuminate the whole negative. That's why still cameras have a minium flash sync time (typically 1/125th). Or am I missing something?

Quote:

Originally Posted by Jason Rodriguez

So if your strobe goes off when integration happens, then you will end up with what you mentioned before . . . part of the frame has integrated, and part of it hasn't. The ProCamHD, since it's a video-based chip, doesn't have a global

Is ProCamHD the chip used in the SI-1920HD?

Quote:

Originally Posted by Jason Rodriguez

reset release, meaning we can't start integration on every line at once. If it did, then the strobe and integration can fire off at the same time, since the entire chip would get exposed at once, and then the rolling shutter would do the read-out after the strobe has fired. Since it is a rolling shutter for both reset and read-out, you have to extend the horizontal blanking, so basically the chip resets every line and integration is happening simultaneously across the entire chip, you fire the strobe, and then the rolling shutter does the read-out. At least this is the way I understand things to happen. There may

If I understand you, the idea is to reset the chip after each frame, so it starts integration from scratch, basically never doing any 'interleaved'/rolling integration? Any idea how many frames/sec. one should expect from driving the sensor in that way?

I am trying to get as much information in advance before calling people with about 8 hours time difference, but I will talk to Steve once I have a rough idea what I am talking about. :)

Thanks very much, Jason!
Martin

Jason Rodriguez · May 30th, 2007, 08:05 AM

The line exposure time can equal the frame time in the electronic rolling shutter unlike a curtain shutter in a film camera which would be physically impossible.

So basically we can set the chip so that it's resetting a line and integrating immediately. Again, the reset point will travel down the length of the chip in 1/60th of a second, but the point is that once a line is reset, it will begin integrating.

The trick will be to make sure your strobe is in-sync with the beginning of a frame-time, not when the reset line is in the middle of the frame. That way you get your entire exposure in a given frame-time and you wont' see the effects of the rolling shutter (basically the row-reset will be in a blanking area before the physical frame integration, what we call the front-porch, so that your strobe goes off and is completed before you start resetting the actual physical frame. That way the entire frame captures the strobe.)

You can definitely talk with Steve about more ideas on this . . . and they may correct me on a thing or two as well :)

Ari Presler · May 30th, 2007, 08:42 AM

In a Rolling Shutter imager, the pixels are ALWAYS integrating light. The pixels can be cleared or reset, but immediately begin integrating again. When an image is readout, it is done on a row-by-row basis, until the entire imager is scanned (Row 1 is readout, then row 2, then row 3 until you roll down to the bottom and start again).

During the row readout, the accumulated value is reset, to start the exposure cycle. It is only during this short time period that light is not valid. If you strobe randomly you would see one row of the image darker than the rest. If you strobe at the top of frame, during the blanking period, the row is not in the visible part of the image.

The time between readout of row-1 and the next time you readout row-1 is the frame time and the maximum possible exposure time. If your frame rate is 60FPS, your max exposure is 1/60sec.

As stated previously, we provide a trigger signal for the strobe at Row-1 (in the non-visible balnking period). Actuall image data usaully doesnt start until line 32.

Martin Breidt · June 6th, 2007, 08:11 AM

Sorry to bother you once more, but we have the additional problem that an integration time of 1/60th is problematic, as we have additional (permanent) light in the scene that we do not want. This is due to the additional components we use in the setup and we cannot turn off the additonal light sources quickly enough for each exposure.

We need the strobe flashes to be the major illumination source for the color camera, so I guess that means the cannot integrate the entire 1/60th but need to go shorter.

Do you still think this is possible?

Thanks very much for yourt help!
Martin

Jason Rodriguez · June 6th, 2007, 08:55 AM

Readout of the sensor takes 1/60th of a second (rolling shutter), so while the line integration time can be shorter (so say an electronic shutter of 1/120th of a second), that's simply the line exposure time changing. The actual read-out speed of the readout of the top line from the bottom line is still 1/60th of a second apart, and that's depenent on the pixel clock of the sensor, so it can't really go faster.

May 23rd, 2007, 10:23 AM	#1
Martin Breidt New Boot Join Date: May 2007 Location: Germany Posts: 11	SI-1920HD and strobe flash I am thinking of integrating an SI-1920HD camera into our computer-controlled recording setup but I am unsure about how the rolling shutter of the camera will handle strobe flash illumination. We are planning on running the camera at 60 frames/sec and will have an external trigger signal for the camera and the strobe, so the strobe will be in sync with the camera. Having read about the camera's rolling shutter and how that means that each sensor line will start integrating at a slightly different point of time, I wonder whether there will be any point in time at which all parts of the sensor will collect light, at which the flash should be then triggered. Otherwise, I might end up with parts of the image properly exposed due to the flash, and parts of the image dark as the flash is already gone or has not yet fired. Any insights much appreciated! Thanks in advance Martin

May 26th, 2007, 01:05 PM	#5
Ari Presler Silicon Imaging Join Date: Apr 2006 Location: New York or Hollywood Posts: 214	Strobe Integration Support Martin, We can simply provide you a strobe output signal from our cameras every top of frame. Set the exposure to full frame time (1/60-sec), which means all rows are integrating light, you will then have correctly exposed images.

May 30th, 2007, 08:42 AM	#9
Ari Presler Silicon Imaging Join Date: Apr 2006 Location: New York or Hollywood Posts: 214	Strobe at top of frame In a Rolling Shutter imager, the pixels are ALWAYS integrating light. The pixels can be cleared or reset, but immediately begin integrating again. When an image is readout, it is done on a row-by-row basis, until the entire imager is scanned (Row 1 is readout, then row 2, then row 3 until you roll down to the bottom and start again). During the row readout, the accumulated value is reset, to start the exposure cycle. It is only during this short time period that light is not valid. If you strobe randomly you would see one row of the image darker than the rest. If you strobe at the top of frame, during the blanking period, the row is not in the visible part of the image. The time between readout of row-1 and the next time you readout row-1 is the frame time and the maximum possible exposure time. If your frame rate is 60FPS, your max exposure is 1/60sec. As stated previously, we provide a trigger signal for the strobe at Row-1 (in the non-visible balnking period). Actuall image data usaully doesnt start until line 32.

May 23rd, 2007, 01:07 PM	#2
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	This can be done, but we would have to activate a special integration mode which would trigger the start of integration while the strobe is off, and not trigger the strobe until the entire frame is integrating, and then read-out the rest of the frame in the dark. I'm thinking though this would extend the frame-time beyond 1/60th of a second . . . but the chip can actually clock a bit higher, so depending on the length of exposure you need, I don't see this being impossible. Since it is a custom application though that would require custom firmware and software to be written, it would be a dedicated development effort, so you should probably contact Steve Nordhauser at (518) 279-9098 and discuss your needs with him to see what can realistically be done.

May 25th, 2007, 11:44 PM	#4
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	Basically the rolling shutter is very similiar to a focal plane array shutter on a still film camera, except that the rolling shutter on a CMOS moves from top->bottom rather than horizontally like on a film camera. So if your strobe goes off when integration happens, then you will end up with what you mentioned before . . . part of the frame has integrated, and part of it hasn't. The ProCamHD, since it's a video-based chip, doesn't have a global reset release, meaning we can't start integration on every line at once. If it did, then the strobe and integration can fire off at the same time, since the entire chip would get exposed at once, and then the rolling shutter would do the read-out after the strobe has fired. Since it is a rolling shutter for both reset and read-out, you have to extend the horizontal blanking, so basically the chip resets every line and integration is happening simultaneously across the entire chip, you fire the strobe, and then the rolling shutter does the read-out. At least this is the way I understand things to happen. There may be some other tricks that can be done in firmware that I don't know about, but again, this would involve custom development, and Steve Nordhauser would be the best one to talk to about that at (518) 279-9098.

May 30th, 2007, 08:05 AM	#8
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	The line exposure time can equal the frame time in the electronic rolling shutter unlike a curtain shutter in a film camera which would be physically impossible. So basically we can set the chip so that it's resetting a line and integrating immediately. Again, the reset point will travel down the length of the chip in 1/60th of a second, but the point is that once a line is reset, it will begin integrating. The trick will be to make sure your strobe is in-sync with the beginning of a frame-time, not when the reset line is in the middle of the frame. That way you get your entire exposure in a given frame-time and you wont' see the effects of the rolling shutter (basically the row-reset will be in a blanking area before the physical frame integration, what we call the front-porch, so that your strobe goes off and is completed before you start resetting the actual physical frame. That way the entire frame captures the strobe.) You can definitely talk with Steve about more ideas on this . . . and they may correct me on a thing or two as well :)

June 6th, 2007, 08:11 AM	#10
Martin Breidt New Boot Join Date: May 2007 Location: Germany Posts: 11	Sorry to bother you once more, but we have the additional problem that an integration time of 1/60th is problematic, as we have additional (permanent) light in the scene that we do not want. This is due to the additional components we use in the setup and we cannot turn off the additonal light sources quickly enough for each exposure. We need the strobe flashes to be the major illumination source for the color camera, so I guess that means the cannot integrate the entire 1/60th but need to go shorter. Do you still think this is possible? Thanks very much for yourt help! Martin

June 6th, 2007, 08:55 AM	#11
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	Readout of the sensor takes 1/60th of a second (rolling shutter), so while the line integration time can be shorter (so say an electronic shutter of 1/120th of a second), that's simply the line exposure time changing. The actual read-out speed of the readout of the top line from the bottom line is still 1/60th of a second apart, and that's depenent on the pixel clock of the sensor, so it can't really go faster.