Human motion detection and tracking are very important research areas in vision based video processing. This thesis presents our efforts to understand and improve human motion detection and tracking techniques. We examined the existing motion detection approaches and proposed our own methods. The first is entropy based motion detection, where the entropy of accumulated difference image is used to detect motion regions. Another simple yet effective motion detection method calculates the optimum threshold for every difference image. Single camera tracking follows motion detection and a mean shift tracking algorithm is firstly implemented. Its advantages and limitations are analyzed and possible improvements are given. We then proposed a single camera tracking method by corresponding detected motion regions. Experimental results demonstrate the effectiveness of our algorithm. A motion detection method based affine tracking method is also presented. Finally, some multi-camera tracking problems are addressed. A homography based method is used for cross camera correspondence. Given multiple cameras, more information is available for the same person. Some preliminary results on selecting the front view of a walking person is shown. Besides, a literature survey on the related areas is also included in the thesis.