Human tracking by adaptive Kalman filtering and multiple kernels tracking with projected gradients

Abstract

Kernel based trackers have been proven to be a promising approach in video object tracking. The use of single kernel often suffers from occlusion since the visual information is not sufficient for kernel usage. Hence, multiple inter-related kernels have been utilized for tracking in complicated scenarios. This paper embeds the multiple kernels tracking into a Kalman filtering-based tracking system, which uses Kalman prediction as the initial position for the multiple kernels tracking, and applies the result of the latter as the measurement to the Kalman update. The state transition and noise covariance matrices used in Kalman filter are also dynamically updated by the output of multiple kernels tracking. Several simulation results have been done to show the robustness of the proposed system which can successfully track all the video objects under occlusion.