Computer Vision evaluate classifier on sub-patches from time t to t+1 actual object position search Region analyze map and update create confidence map set new object classifier position (tracker) - - Self-learning! + - -
Computer Drifting Vision Tracked Patches Confidence
Computer Drifting Vision
Computer Vision Tracking by Detection Detection (object class)
Computer Traditional Tracking Vision t=1 t=2 position in prev. frame initialization candidate new positions (e.g., dynamics) best new position (e.g., max color similarity)
Computer Tracking-by-Detection Vision … detect object(s) independently in each frame associate detections over time into tracks
Computer Multiple Objects Vision Frame 1 Frame 5 Frame 9
Example: Multiple Object Computer Vision Tracking
Computer How to get the detections? Vision Persons Background Supervised Learning
Computer Using the classifier Vision
Computer How to link them? Vision • Space-Time Analysis: (a) collect detections Space Time Volume Detections [Leibe et al. CVPR’07]
Computer Trajectory Estimation Vision (a) collect detections (b) trajectory growing and selection t t z x Space Time Volume
Computer Trajectory Estimation Vision (a) collect detections (b) trajectory growing and selection t t H 2 H 1 z x Space Time Volume
Computer Result Vision Input (Object Detections) “Tracking” Result
Computer Vision
Recommend
More recommend
