digital image stabilisation as a gimbal alternative
-
Hello everyone,
is it possible to digitally stabilise the video image in real time with the VOXL2?
I am aware that image information is lost at the edges. One idea is to stabilise a 4k video image and then only transmit a stabilised HD video section from the centre.
Here you can see what I mean:
https://github.com/AdamSpannbauer/python_video_stabI just wanted to show with the link what I mean. You would probably need other software that can also use hardware acceleration on the VOXL2.
Best regards
Tom -
Hello @TomP ,
The digital image stabilization is on our road map, but we don't have any official dates at this time.
The GPU on Voxl is very powerful and we are working on some features that make use of zero-copy approaches and optimized OpenCL functions (for Adreno GPU) for manipulating / transforming pixels in an image, filtering, undistortion, etc. We also hope to provide some examples of optimized image manipulation on GPU sooner rather than later.
VOXL2 is powerful enough to do full 4K30 (at least) stabilization in real-time. In order to achieve that, we can select the full frame resolution on the camera, which is 4056x3040 (for IMX412 camera) and we have a little wiggle room horizontally and a lot of wiggle room vertically (margin for stabilization). However, the margin can be increased by applying digital zoom, so the output image would be automatically up-sampled to 4K from a slightly smaller stabilized region (let's say 80% smaller). A smaller resolution image (lets say 2K or 1080p) can also easily be generated (the zoom / scale factor does not have to be an integer).
The stabilization algorithm will also make use of the available IMU on VOXL2.
Are you interesting in developing your own digital stabilization solution or using our (future) solution?
Alex