A new joint sequential object detection and tracking approach and its performance analysis

Object detection and tracking are two important research topics in surveillance systems. Object detection and tracking are typically implemented separately and object tracking is usually performed after the object is detected. This two-stage approach may not work well when the object has a very low signal to noise ratio (SNR), and cannot be reliably detected using a single sample. In this paper, a new joint sequential object detection and tracking algorithm which combines Wald's sequential probability ratio test (SPRT) and the Kalman filter is proposed. Theoretical results have been provided on the expected values of the test statistic under both hypotheses, to give insights on the termination of the SPRT procedure. Numerical results show that even with a very weak SNR, the average number of samples required by the proposed sequential detector to achieve a high detection performance is small, and that it needs a much smaller number of samples to achieve the same detection performance than the optimal fixed-sample-size (FSS) detector.