Output details

Occlusion analysis: learning and utilising depth maps in object tracking

<23> One of the earliest reported works on recovery and utilisation of scene structure in visual surveillance detection and tracking algorithms. The key innovation was the use of moving objects in the scene as probes for depth. A height model (itself parameterised from scene activity) allowed the "distance" from camera (actually a pixel-based proxy for this depth) to be inferred. The availability of this scene structure enabled the tracking algorithm to perform accurately where the scene exhibited significant static occlusions i.e. typical public space scenery.