A Sparse Probabilistic Learning Algorithm for Real-Time Tracking

Proc. Int. Conf. on Computer Vision |

This paper addresses the problem of applying powerful pattern recognition algorithms based on kernels to efficient visual tracking. Recently Avidan [1] has shown that object recognizers using kernel-SVMs can be elegantly adapted to localization by means of spatial perturbation of the SVM, using optic flow. Whereas Avidan’s SVM applies to each frame of a video independently of other frames, the benefits of temporal fusion of data are well known. This issue is addressed here by using a fully probabilistic ‘Relevance Vector Machine’ (RVM) to generate observations with Gaussian distributions that can be fused over time. To improve performance further, rather than adapting a recognizer, we build a localizer directly using the regression form of the RVM. A classification SVM is used in tandem, for object verification, and this provides the capability of automatic initialization and recovery. The approach is demonstrated in real-time face and vehicle tracking systems. The ‘sparsity’ of the RVMs means that only a fraction of CPU time is required to track at frame rate. Tracker output is demonstrated in a camera management task in which zoom and pan are controlled in response to speaker/vehicle position and orientation, over an extended period. The advantages of temporal fusion in this system are demonstrated