Robust Methods for Visual Tracking and Model Alignment
MetadataShow full item record
The ubiquitous presence of cameras and camera networks needs the development of robust visual analytics algorithms. As the building block of many video visual surveillance tasks, a robust visual tracking algorithm plays an important role in achieving the goal of automatic and robust surveillance. In practice, it is critical to know when and where the tracking algorithm fails so that remedial measures can be taken to resume tracking. We propose a novel performance evaluation strategy for tracking systems using a time-reversed Markov chain. We also present a novel bidirectional tracker to achieve better robustness. Instead of looking only forward in the time domain, we incorporate both forward and backward processing of video frames using a time-reversibility constraint. When the objects of interest in surveillance applications have relatively stable structures, the parameterized shape model of objects can be usually built or learned from sample images, which allows us to perform more accurate tracking. We present a machine learning method to learn a scoring function without local extrema to guide the gradient descent/accent algorithm and find the optimal parameters of the shape model. These algorithms greatly improve the robustness of video analysis systems in practice.