Video Signal Process In Network Camera Based On DSP Platform

Along with the development of the society and technology, video surveillance has come to the digitalized developing period. The appearance of the high-resolution network camera results in the great achievement in video surveillance area. The core technology is how to realize the high efficient video signal process in the system, such as video signal compress, denoising, γ correction and so on. As a result, the paper mainly studies the design and the optimization of the H.264 encoder and some preprocess algorithms of video signal.At first, the paper analyzes the noise in video signal and studies the models of noise. A fast noise estimation algorithm is proposed, which makes the denoising process adaptive and affects the threshold of some algorithm in encoding process.Secondly, the paper analyzes the core tools in H.264 encoder. By evaluating the complexity and the performance together, the encoder is optimized to satisfy the application in network camera. The edge and texture information is used to evaluate the best prediction mode for 4*4 blocks in intra prediction encoding process. With the decided 4*4 block mode information, the most probable 16*16 block mode is predicted. Compared to the full search algorithm, the proposed algorithm reduce 30% to 70% computations.Thirdly, algorithms are optimized on DM642 platform. By fully making use of the DM642 CPU structure and instruction, linear assembly program language is used to improve the key functions which execute frequently and constitute the main complexity. This method reduces the executing time of every key function by 50% to 70%, then the encoder efficiency is improved. Cache optimization also bring great improvement in efficiency which is based on the 2 level cache structure in DM642. CPU stall is avoided when EDMA controller is used to move data for CPU process from external sdram to on chip sram in parallel with CPU processing. By avoiding extra data transfer between SDRAM and SRAM, data transfer strategy is optimized, for example, the optimized process strategy of Bayer color pattern data work 2 times faster than before. The algorithms performance is improved greatly in this step.At last, the encoder is integrated into the network camera system which is based on RF5. Important modules are designed and algorithms are encapsulated, then the H.264 encoder works well in the system.