Long Exposures

[IanCassTwo]

Hi,

I'm writing a driver for astrophotography and whilst testing it, I notice that the exposure lengths are wrong. I'm running repeated 60 sec exposures. From a cold start, the exposure times from triggering an async capture_array to the job returning (with a wait) is as follows (in seconds):- 1, 61, 119, 119, 119, 119, etc... Not one of them is the 60 seconds that I asked for. In fact, most take almost 2mins! Any thoughts why?

Some code snippets below to show how I'm doing this.

config = picam2.create_still_configuration( {"size": (640, 480)}, raw={'format': 'SRGGB12','size': (2028,1520)})
picam2.configure(config)
picam2.start()
...
with picam2.controls as controls:
controls.ExposureTime = int(duration * 1e6)
controls.AeEnable = False
controls.NoiseReductionMode = libcamera.controls.draft.NoiseReductionModeEnum.Off
controls.AwbEnable = False
controls.AnalogueGain = gainvalue
controls.ColourGains = (2.0, 2.0)

job = picam2.capture_array("raw", signal_function=oncapturefinished)
...
array = picam2.wait(job).view(np.uint16) * (2 ** (16 - 12))

EDIT:-

If I specify "queue=False, buffer_count=2" in the configuraton, I get different results. This time, for a 60 sec exposure, it would take 1 sec for the first exposure, then each subsequent one would take 58 seconds. Likewise, for a 2min exposure, the first takes 1 second and the rest take 1:59secs. What's happening here?

[davidplowman]

Hi and sorry for the delay.

There might be a few different things happening. Firstly, when you set the exposure time it can take a while for it to take effect. If you can, maybe try setting things up before starting the camera (picam2.start()). Also note that still configurations (as created by create_still_configuration) default to only 1 buffer (because the images are usually large and some Pis would run out of memory). This means you are guaranteed to drop at least every other frame. So buffer_count=2 (or higher!) is almost certainly a better choice. If you're measuring the capture-to-capture time, rather than the reported exposure on the images, that would account for those 2 minute delays.

[IanCassTwo]

Thanks for the answer. With buffer_count=2 I'm still seeing the odd thing where a 60sec exposure will be returned in less than 60 secs. I suspect it's because the camera is filling the next buffer whilst I'm still processing the previous one, therefore there's some overlap due to processing time. Is this likely the case?

[davidplowman]

How are you measuring the exposure time? It's possible that the image is already being exposed when you start to wait for it. I would check the 'ExposureTime' field in the image metadata as this should always be accurate and correct, perhaps something like this:

request = picam2.capture_request()
array = request.make_array('raw')
exposure_time = request.get_metadata()['ExposureTime']
request.release # don't forget this!!

[IanCassTwo]

I'd be measuring from start of exposure (capture_array) to the time the callback gets triggered. I trigger the exposure like this.

state.job = picam2.capture_array("raw", signal_function=oncapturefinished)

then fetch the result like this:-

array = picam2.wait(state.job).view(np.uint16) * np.left_shift(1, (16 - 12))

How do I also ask for metadata?

[davidplowman]

I'd capture the whole request rather than just the image. If you need it to be non-blocking, you can call picam2.capture_request(signal_function=oncapturefinished), and then picam2.wait(job) will give you the request and you can copy what I posted just above.

[IanCassTwo]

Perfect, thanks. I'll give it a try

[IanCassTwo]

Confirmed, the image is returning ~57 secs after exposure started, but the image metadata says exposure is 59999485. It looks like the camera started exposing the next frame in the background even before I triggered it probably because I told it I wanted 2 frames in the buffer. That behaviour suits my application quite well.