The FPGA Implementation Of Visual Tracking Algorithm Based On Particle Filter

Visual tracking is one of the key issues in the field of computer vision,it has been widely used in video surveillance,interactive, sports analysis, visual navigation and military identification areas.Among many algorithms, Particle Filter has been widely used with the advantage of solving nonlinear and non-Gaussian problems. In this essay, the algorithm can adapt to the situation of target size change, target rotation, light change and partial occlusion, so the robustness of the algorithm is relatively strong, but due to the large number of particles need too much operation, the real-time characteristic is unable to meet the requirement. Based on this, this easy developed a FPGA based hardware algorithm, which can be 20 times faster than the software algorithm and it can fully meet the requirements. This essay first studies the principles of particle filter algorithm which can be a basis for the follow-up studies. The main work and research results are as follows: 1. Use Open CV to simulate the particle filter algorithm. The result shows it can adapt to situation of target size change, target rotation, light change and partial occlusion. 2. Implement the algorithm on FPGA including the SDRAM controller, Gaussian generator, particle transitions, particle sorting, particle histogram, particle weight calculation and resampling module. 3. Simulate the hardware algorithm in Modelsim and build a demonstration system which is based on DE2-115 toolkit. The system uses PAL video as input, then do the video capture and particle filter processing in the DE2-115 platform, at last display the tracking results on the VGA monitor. The system can track targets in real time.