Video annotation for computer vision is the process of labeling objects, actions, or regions in video frames to create ground-truth data for computer vision models. It involves drawing bounding boxes, polygons, segmentation masks, or keypoints on objects of interest in each frame.