Realtime Detection and Tracking

The Vision Group at the CAMP chair is working i.a. on real-time detection and tracking of arbitrary objects. Tracking is the process of finding the pose of an object relative to the camera. To fulfill this task, a priori information can be taken into account (Tracking) or dropped (Tracking-by-Detection). The former method results in a more efficient performance but fails if the object movement is too big. Furthermore, it can not reinitialized again if the algorithm loses track. Contrary to dealing with a priori information, Tracking-by-Detection is slower but performs more general. That means that it tries to find an object in the whole image without any previous knowledge where the object could be. Therefore, Tracking-by-Detection is well suited for (re-)initializing Tracking methods. The combination of both approaches optimizes the performance of object tracking with respect to speed and robustness.

The computer vision group at the CAMP chair follows both paths where it investigates different object features like texture, edges and feature points. The work of the group has gained worldwide reputation on many international conferences.