Foreground Classification Basing On Active Features For Visual Surveillance

The subject is aimed at the technology of moving target recognition in traffic scenes, which is one of the key technology of intelligent video surveillance system. At present, ITS is mostly focused on the detection and recognition of vehicle, and it is either considered without pedestrian and bicycle/motorcycle or focused a little on them.The recognition and classification of vehicle, bicycle/motorcycle and pedestrian is a complex process. The normal template matching algorithm for recognition needs a lot of computations, so that it can hardly satisfy the real-time and effective requirements in surveillance. To solve this question, we presents an integrated framework integrating active template matching algorithm and context information to recognize the moving target. This framework can be an independent module embedded into a visual surveillance system.The active templates are in the form of a dictionary of active haar features (bases), which are allowed to slightly shift at different locations and orientations. They can be learned for each object type from a small set of positive samples that roughly aligned. With these learned deformable templates, the moving foregrounds subtracted from background model are recognized through searching maximum matching likelihood. To avoid the exhaustive search for template matching and reduce the noise disturbance, a scheme to estimate target size and pose at specific location is developed based on the contextual information of scene geometry and tracking cues. This framework can be an independent module embedded into a visual surveillance system. Its performance and benefit of using context are quantitatively demonstrated on public dataset with comparisons. Experiments show this framework can work well and satisfy the real-time requirement.We recognize the targets in traffic scene as three categories, pedestrian, sedan and bicycle/motorcycle. Experiments are performed in different scenes. We compare our results with the results of Adaboost algorithm quantitatively. Its performance and benefit of basing on active template matching and using context information are well demonstrated with comparisons.