Abstract:

The particle filter is a popular tool for visual tracking. Usually, the appearance model is either fixed or rapidly changing and the motion model is simply a random walk with fixed noise variance. Also, the number of particles used is typically fixed. All these factors make the visual tracker unstable. To stabilize the tracker, we propose the following measures: an observation model arising from an adaptive noise variance, and adaptive number of particles. The adaptive-velocity is computed via a first-order linear predictor using the previous particle configuration. Tracking under occlusion is accomplished using robust statistics. Experimental results on tracking visual objects in long video sequences such as vehicles, tank, and human faces demonstrate the effectiveness and robustness of our algorithm.