Adaptive Motion Estimation Algorithm Based On Multi-Minimum Particle Swarm Optimization

There are a lot of time redundancy in the video image, using motion estimation can reduce the correlation between frames, and then improve the coding efficiency. Motion estimation is more accurate, more high quality of the image compensation. Because the block matching motion estimation algorithm is easy to implement in software and hardware, so it is a hot research in the field of video compression coding. The full search algorithm, which belongs to the most primitive block matching motion estimation algorithm, its search accuracy is the highest, but the huge amount of calculation is difficult to achieve real-time application. Almost all the existing fast motion estimation algorithm is proposed according to the hypthesis that the error function is unimodal, although they can shorten the search time and improve the convergence ability of algorithm, but the motion estimation algorithm which can not achieve a breakthrough in essence still exist local optimal, these algorithms can not both search precision and search efficiency, so it is difficult to meet the actual application requirements.First of all, this thesis analyzes the principle of block matching motion estimation, introduces several classic motion estimation algorithms, and then introduces the improved particle swarm optimization algorithm, analysis the performance of the algorithm by experiment data, finally based on the disadvantages of the existing algorithm, this paper puts forward a kind of based on minimum fast motion estimation algorithm of particle swarm optimization. The algorithm uses the globally search ability of multiple minimum particle swarm algorithm to solve motion estimation, it can adapt to the condition that error function contains multiple minimum. The algorithm through the centerbiased characteristics of motion vector and the randomness of the particles set original particles, use adaptive exercise intensity for different video sequences, set up different number of iterations and thresholds in the process of particle iteration.On Matlab simulation platform, the algorithm is compared with other similar algorithms by five different motion degree of standard video sequence. The experimental results show that the computational complexity of algorithm is similar to the diamond algorithm for the video sequences with slow and middle motion. For the video sequences with violent motion, the computational complexity of algorithm is close to the threestep search algorithm. In the case of increasing a few search points, the search accuracy was close to the full search algorithm for all kinds of video sequences. Compared with similar algorithms, the algorithms not only can meet the real-time requirement of video decoding by fast convergence, but also can obtain high precision.