Hackacam. Programmable 1080p30 H.264 camera module with interchangeable lens

This looks great!

I'm interested in using this for aerial search applications. At the moment we use a PtGrey Chameleon 1280x960 camera, connected over USB to a pandaboard. We run python/C image recognition code on the pandaboard. See http://canberrauav.com/ for details of our project.

There are a number of things that are appealing about your project. First off, we could potentially do the image recognition directly on the ARM9 on the camera, instead of on the pandaboard. To do that we'd really need to take advantage of the DSP, as right now we're stretching the ARM on the pandaboard quite a lot to get it to do the recognition we want. We capture raw bayer grid frames at 7.5 fps.

A few questions if you have time:

we want to do realtime recognition of the images at around 7 to 10 fps, but we also want to save all the images in a lossless format to storage so we can replay a flight and try new recognition algorithms on the same data. How much bandwidth do you think we can get to an SD card on the camera to save lossless images? Will writing to the SD card tend to clag up the camera for the recognition task?
how sensitive is the sensor? A major reason for choosing the chameleon is it has a very sensitive sensor which means very small frame capture times (typically less than 100 microseconds). This means vibration in the plane doesn't cause pixels to be smeared.
we need good time synchronization between camera frames and the APM attitude calculation. This means we need to know exactly when each frame was captured, and need to be able to set/query the timebase of the camera vs the timebase of our other onboard computers (APM and pandaboard). How would we do that?
we need to be able to control the brightness/gain of the camera fairly carefully, so that small bright objects have plenty of dynamic range. This means we usually want the overall image fairly dark, so that small bright objects don't get washed out. Can we control that? (we set the gamma and brightness/gain in the chameleon based on the last image to do this now)
how heavy will the camera be?
how is it powered, and how much current will it draw, at what voltage?

If you are interested, our current image recognition code is here:

https://github.com/tridge/cuav/blob/master/image/scanner.c

and the code we use to control the chameleon is here:

https://github.com/tridge/cuav/tree/master/camera

It's basically low level C code providing a python interface.

Cheers, Tridge

CanberraUAV

Open Source Civilian UAV Development

Replies

Developer

John Arne Birkeland December 18, 2012 at 1:40am

I pressed the pledge button before I even finished reading the kickstarter page..
Developer
Andrew Tridgell December 18, 2012 at 12:50am
This looks great!

I'm interested in using this for aerial search applications. At the moment we use a PtGrey Chameleon 1280x960 camera, connected over USB to a pandaboard. We run python/C image recognition code on the pandaboard. See http://canberrauav.com/ for details of our project.

There are a number of things that are appealing about your project. First off, we could potentially do the image recognition directly on the ARM9 on the camera, instead of on the pandaboard. To do that we'd really need to take advantage of the DSP, as right now we're stretching the ARM on the pandaboard quite a lot to get it to do the recognition we want. We capture raw bayer grid frames at 7.5 fps.

A few questions if you have time:
- we want to do realtime recognition of the images at around 7 to 10 fps, but we also want to save all the images in a lossless format to storage so we can replay a flight and try new recognition algorithms on the same data. How much bandwidth do you think we can get to an SD card on the camera to save lossless images? Will writing to the SD card tend to clag up the camera for the recognition task?
- how sensitive is the sensor? A major reason for choosing the chameleon is it has a very sensitive sensor which means very small frame capture times (typically less than 100 microseconds). This means vibration in the plane doesn't cause pixels to be smeared.
- we need good time synchronization between camera frames and the APM attitude calculation. This means we need to know exactly when each frame was captured, and need to be able to set/query the timebase of the camera vs the timebase of our other onboard computers (APM and pandaboard). How would we do that?
- we need to be able to control the brightness/gain of the camera fairly carefully, so that small bright objects have plenty of dynamic range. This means we usually want the overall image fairly dark, so that small bright objects don't get washed out. Can we control that? (we set the gamma and brightness/gain in the chameleon based on the last image to do this now)
- how heavy will the camera be?
- how is it powered, and how much current will it draw, at what voltage?
If you are interested, our current image recognition code is here:

https://github.com/tridge/cuav/blob/master/image/scanner.c

and the code we use to control the chameleon is here:

https://github.com/tridge/cuav/tree/master/camera

It's basically low level C code providing a python interface.

Cheers, Tridge
CanberraUAV

Open Source Civilian UAV Development

This reply was deleted.

Discussions

Hackacam. Programmable 1080p30 H.264 camera module with interchangeable lens

Replies

Activity