A little script I wrote in Python 2.7 and OpenCV 3 to make a still image appear like a video recording.
The aim of the program is to keep the output reasonably realistic while also performing data augmentation to get a seemingly limitless video feed.
- Randomized image panning
- Randomized image rotation
- Occasional defocusing with slight zoom (note that true bokeh blurring is technically impossible to simulate, but Gaussian blurring is a decent approximation of it, so I've used that instead)
- White balance (brightness adjusts as the image moves)
- Motion blurring (again, a very rough implementation, since a realistic one is hard to simulate with just a single image)
- All of these are parametized, so it's very easy to tweak the output to your liking. I haven't experimented with the parameters much myself, but you should be able to get a pretty realistic output.
Sample video This example shows the program in action with a ROS wrapper that takes a picture from the webcam (using cv_camera, not shown in the video) and displays a video output.
Pretty straightforward. Pass the file name as a parameter: python augment.py filename.jpg
To exit, hit q.