LukeAI
commented
5 years ago you would first need some way of associating bounding boxes between frames. hungarian association is a simple efficient way to do that. then you can interpolate the bounding box corners. simple linear interpolation could be a good starting point, just to see if things are roughly working. spline interpolation would be better than linear because it takes into account more than just 2 frames. https://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.interp2d.html
Hi, I would ask whether there is an algorithm to do the interpolation. When I use yolov3 to detect object in the video, I framed video to get images first, so is it possible to just detect frame 1 and frame 5 and use bbox information to do interpolation on frame 2 3 4 rather than use yolov3 model to detect on them.