Study And Development On Intelligent Video Surveillance System Based On H.264

The network video surveillance system has become the mainstream in the field of video surveillance. But in the applications, there exist some problems for network video surveillance system to be solved, such as the misjudgment of the security situation due to the visual fatigue of security guards, the conflict between the requirement for the quality and the real-time performance under the limited bandwidth, the high cost of the system which is burdensome to those small and medium-sized enterprises. Based on the DSP, the intelligent surveillance technology and the latest video codec standard H.264 are combined to construct a video surveillance system in this thesis. The main achievements and innovations are described as follows,1. To solve the error accumulation existing in the algorithm of initializing the background based on the pixel intensity classification, the sequence is divided into several steady sections according to the temporal correlation between the adjacent frames. Then the steady sections are merged into the consistent sections based on the mean value of the members. In addition, a weighted approach based on shift operator which is more appropriate for DSP is proposed to upgrade the background. It is demonstrated that the scene change detection method in this thesis guarantees the accuracy and the real-time performance to detect the moving object.2. An algorithm of mode decision and motion estimation is proposed based on the result of the scene change detection. Every frame is partitioned into four regions including the background region, the active region, the overlapping region and the shading region which are processed with different reduced sets of encoding modes. The motion vector of the active region is initialized by the relative position between the current frame and the reference. Experimental results show that the proposed algorithm performs 30% time-saving than x264 with a little quality descending.3. The modeling method for dynamic relationship between the quantization parameter and encoding complexity is presented based on the least squares support vector regression. The constant bit rate control approach realized by means of the dynamic model is applied to improve the rate control performance of H.264 under the low bandwidth. The rate control error of the proposed algorithm is effectively reduced in average with a more stationary quality than x264 in several tests of different sequences.4. The intelligent video surveillance terminal based on the high performance digital signal processor TMS320DM642 is designed according to eXpressDSP which is known as the advanced software design concept of DSP. An algorithm to optimize SATD module investigated on the basis of pipelines and the technoly of single instruction multiple data (SIMD) can reduce the computational complexity by means of parallel processing. Experimental results show that the accuracy of the moving objects detecting and the real-time performance of H.264 encoding can meet the demand in the application.