Tracking with dynamic weighted compressive model

• An online feature selection algorithm with dynamic update random matrix is proposed.
• Adopt multiple matrices with different dimensions to deal with the scale problem.
• A weighting scheme is used to evaluate the performance of each matrices.
• A feedback mechanism is employed to reduce the tracking drift problem.

Fast compressive tracking utilizes a very sparse measurement matrix to capture the appearance model of targets. Such model performs well when the tracked targets are well defined. However, when the targets are low-grain, low-resolution, or small, a single fixed size sparse measurement matrix is not sufficient enough to preserve the image structure of the target. In this work, we propose a multi-sparse measurement matrices scheme along with a weight map to select the best measurement matrix that preserves the image structure of the targets during tracking. The weight map combines a contrast weight and a feature weight to efficiently characterize the target appearance and location. Moreover, a dispersion function is used for the online update of the target template, allowing tracking both the location and scale of the target. Extensive experimental results have demonstrated that the proposed DWCM tracking algorithm outperforms several state-of-the-art tracking algorithms as well as compressive tracker.