The main contributions are summaried as follows: First, a new parallel hardware architecture for Intelligent Video Surveillance system is designed based on PXA270 Xscale processor and multiple TMS320DM642 DSP processors. The designed architecture’s characteristics, hardware configuration, and working principle are elaborated. In addition, Experimental results validated the performance of the proposed system. Second, a moving objects detection algorithm suitable for embedded system is presented, which is based on Mixture of Gaussian theory and fusing spatial-temporal information. In this method, the number of Gaussian function varies with the complexity of background dynamically, which decreased the computational complexity and memory requirement with no loss of detecting accuracy. Besides, a spatial-temporal histogram correcting/confirming algorithm is proposed in order to reduce detection error. Finally, the detection algorithm is testified in DSP system. Third, a small moving objects tracking algorithm suitable for embedded system is designed. The posterior probability of each pixel belonging to the target is estimated and the probability distribution image is created, which based on the theory of Bayesian probability estimation according to the character distribution of object and background. The proposed algorithm is able to enhance the feature which manifests greater difference between the object and the background nearby adaptively. Besides, this probability distribution is also introduced for the updating procedure of target features, so the robustness of the algorithm is greatly improved. Finally, based on the new probability distribution image, as well as by combining with Kalman prediction and Camshift algorithm, the performance of the new proposed algorithm is testified, the experimental results from DSP platform indicate that the proposed algorithm can tracking small object in real-time, and can deal with the occlusion problem effectively. Finally, the strategies of dynamic task scheduling and static task scheduling for multiple cameras are put forward. At the same time, the cooperation mode of detection and tracking is introduced. In the end, the muli-cameras system is used in real environment.
修改评论