New Object Detection, Tracking, and Recognition Approaches for Video Surveillance Over Camera Network

Object detection and tracking are two fundamental tasks in multicamera surveillance. This paper proposes a framework for achieving these tasks in a nonoverlapping multiple camera network. A new object detection algorithm using mean shift (MS) segmentation is introduced, and occluded objects are further separated with the help of depth information derived from stereo vision. The detected objects are then tracked by a new object tracking algorithm using a novel Bayesian Kalman filter with simplified Gaussian mixture (BKF-SGM). It employs a Gaussian mixture (GM) representation of the state and noise densities and a novel direct density simplifying algorithm for avoiding the exponential complexity growth of conventional Kalman filters (KFs) using GM. When coupled with an improved MS tracker, a new BKF-SGM with improved MS algorithm with more robust tracking performance is obtained. Furthermore, a nontraining-based object recognition algorithm is employed to support object tracking over nonoverlapping network. Experimental results show that: 1) the proposed object detection algorithm yields improved segmentation results over conventional object detection methods and 2) the proposed tracking algorithm can successfully handle complex scenarios with good performance and low arithmetic complexity. Moreover, the performance of both nontraining- and training-based object recognition algorithms can be improved using our detection and tracking results as input