Implementation And Optimization Of AVS Video Encoder Based On DSP Platform In The Application Of Video Surveillance System

The technique of video surveillance is one of the most popular researches in the world today. Video compression technique with high efficiency has become the key point of the development of surveillance application. AVS (Advanced Audio Video Coding Standard) is our national digital audio and video encoding and decoding standard with the intellectual property rights owned by China, which has many advantages such as high-performance, low-computing complexity and low-cost of patent fee. TMS320DM6446, a new general digital signal processor, is developed by TI for the application of HDTV. This thesis presents the research of implementation and optimization of AVS video encoder based on DSP platform. The purpose of the research is to realize a real-time encoding system which can be applied in video surveillance. It has significance in the development of AVS.According to the requirements of real-time and high efficiency, this thesis has studied AVS-P2 (Video Encoding) based on recent research of video surveillance. After the optimization on algorithm and flow, the AVS encoder is implemented on DM6446 and optimized further based on DSP features. In the part of algorithm optimization, the most time-consuming module, inter-prediction algorithm module, is the main optimizing point. According to the statistics of inter-prediction modes of those video sequences in different scenarios and then considering the characteristics of the video sequences in video monitoring, an optimized inter-prediction mode decision algorithm for video surveillance is proposed. After analyzing integer-pixel and sub-pixel searching algorithms in motion estimation module, an optimized fast searching algorithm is proposed in order to reduce the complexity of computing and improve the efficiency. Optimization on the flow of inter-prediction is also helpful to raise encoding speed. AVS video encoding speed has already been raised in a high level by algorithm optimization. Implementation and optimization of the encoder according to the features of DSP platform is the following work in this thesis in order to make the system more suitable to the engineering application. The methods of optimization include using complier options, data structure optimization, memory distribution, applying EDMA and CACHE and so on. Linear assembly is an important method to improve performance of pipeline. Considering the complicated algorithm of interpolation, by analyzing the features of algorithm and the target platform characteristics, the interpolation algorithm is proposed by transforming data structure, linear assembly, Intrinsics, data packing etc. All these methods can reduce encoding time significantly but maintain the performance based on DSP platform.Through the research of AVS and its applications, an optimized AVS encoder is achieved to do video monitoring work. The test results show that the optimized AVS video encoder based on DSP platform can meet the requirements of real-time encoding with the negligible drop of performance. Therefore, this AVS encoder can be applied in video surveillance. The achievements of this thesis are very important for the practical application of AVS especially in video surveillance.