Tracking and Clustering Salient Features in Image Sequences

Many applications in media production need information about moving objects in the scene, e.g. insertion of computer-generated objects, association of sound sources to these objects or visualization of object trajectories in broadcasting. We present a GPU accelerated approach for detecting and tracking salient features in image sequences and we propose an algorithm for clustering the obtained feature point trajectories in order to obtain a motion segmentation of the set of feature trajectories. Evaluation results for both the tracking and clustering steps are presented. Finally we discuss the application of the proposed approach for associating audio sources to objects to support audio rendering for virtual sets.