Research Of The Global Motion Estimation Technology In Video Sequences And The DSP Implementation

Moving target detection is the basis of intelligent visual surveillance, and has a wide application prospect in the field of military, traffic, industry, biomedical and so on. Among them, the detection of moving target in dynamic scene still has to face many challenges. At present, the main problem of detection in dynamic scene is susceptible to interference from former sites which makes the accuracy of detection is difficult to ensure, and large amount of calculation leads to the difficulty of real-time detection. This issue limits the widespread adoption of the moving target detection in dynamic scene. Based on the existing research results, and aimed at the main problem of moving target detection in dynamic scene, this paper makes in-depth research.A fast global motion estimation algorithm based on image pyramid decomposition is presented in this paper. In order to improve the calculation speed and the accuracy, a three-tier pyramid method is utilized in the multi-resolution calculation. In the first layer of the pyramid image, the algorithm predicts the blocks, and removes the unreliable blocks. This algorithm uses the block matching algorithm to roughly find the camera motion parameters. And the motion parameters are mapped to the second layer of the pyramid. In each layer image of the pyramid, the parameters are calculated by a double iterative least squares method. This method further improves the accuracy of the calculation. Experiments show that, the speed and the accuracy of the algorithm has improved to some extent.Consider the value that the algorithm implemented on DSP, in this paper, it is applied in DM6437 successfully. We use some methods such as the compiler optimization options in CCS, EDMA data movement, put frequently used data into the chip, software stream line, linearity compilation and so on to optimize the algorithm, and make its running on DM6437 to be real time. Experiment results show that the algorithm has a good performance on the DSP, and the algorithm can meet the real-time requirements.