Object tracking is the task of following one or multiple objects in a video over time, associating detections of the same object across frames. Given an initial detection (sometimes provided, as in tracking-by-detection, or maybe user-initialized), a tracker will update the object’s position (and sometimes other states like size, velocity) in subsequent frames. Tracking can be single-object or multi-object. Multi-object tracking often uses detection results in each frame (like from YOLO or Faster R-CNN) and then performs a data association step to link detections to existing tracks (e.g., via Hungarian algorithm on some distance metric). There are also algorithms like Kalman Filters (with a motion model) for prediction, and optical flow-based trackers (like Lucas-Kanade tracker for feature points, or more modern ones like ECO, SiamFC which use correlation or deep learning to directly track the appearance). Challenges in tracking include handling occlusions (object disappears and reappears), changes in appearance, multiple similar-looking objects, and real-time performance. Applications: surveillance (track people), sports analytics, autonomous driving (track cars/pedestrians), etc.