Robust Multi-feature Visual Tracking with a Saliency-based Target Descriptor

In this article a robust multi-feature tracking algorithm based on visual saliency mechanism and particle filter is proposed in order to avoid the tracking failure of single feature in complex environment. The target template is composed of visual saliency feature and other two features extracted from VISible (VIS) and InfraRed (IR) frames initially; more specifically, compressive sensing method is employed under the particle filtering framework in order to reduce the computation. Meanwhile, this target template is adaptively updated according to Bhattacharyya coefficients. Experimental results demonstrate that this algorithm is more accurately and effectively under conditions of illumination variation, clutter, as well as similar background and occlusions with good robustness than other existing tracking algorithms.