KLT

lastmod 2026-11-10created 2022-07-123 edits

KLT (Kanade-Lucas-Tomasi) is a classic computer vision technique that detects keypoints in an image and tracks them across consecutive frames. It combines estimating keypoint motion with Lucas-Kanade optical flow and choosing points that are good to track with Shi-Tomasi corner detection. In OpenCV it can be implemented by combining goodFeaturesToTrack and calcOpticalFlowPyrLK.

See also

• Feature detection
• Motion estimation