The Python program gs_homography.py computes the coordinates of the 4 corners of the green screen on each video frame in the videos of the multi-object tracking scenario in Armory, assuming accurate green screen coordinates are present on the reference frame of the video.
The reference frame can either be the first frame of the video, or explicity provided as a parameter.
We develop this program because we have observed that the existing green screen coordinates are not stationary in the virtual world of some video frames, causing the green screen (or patch that would be placed on the green screen) to "move."
The program copies the original videos from a user-provided path (i.e., the "old" dataset), to a new location at a path provided the user (we call this the "new" dataset), and computes and stores the computed green screen coordinates in the new dataset as calculated according to the parameters provided in homography.json.
The program also visualizes the green screen coordinates of the old data as red dots and new data as yellow dots. The video frames with the visualized coordinates are stored in the rgb_w_coords subfolder within each video folder at the new dataset's path. Below we show an example of how the coordinates may be corrected by the program ("before" in red vs "after" in yellow).
python3 gs_homography.py {json file}While running the program it is necessary to ensure the reference frames from old_data have accurate green screen coordinates