Research On Video Analysis Methods Based On Improved ViBe And Particle Filter

With the rapid development of internet of things, big data and other modern information technology, moving target detection and tracking as the basis for the field of machine vision technology plays an increasingly important role in intelligent video surveillance. Many scholars have done a lot of research in this field and achieved fruitful results. However, due to the complexity of the requirements and application scenarios, such as changes in lighting, noise and object occlusion problems, moving target detection and tracking technology has long been challenges in the field. In this thesis, the current mainstream methods of moving target detection and tracking are studied deeply and more robust and faster improved algorithms are proposed for solving shortcomings of current mainstream methods. After full analysis of experiments, the effectiveness of improved algorithms is verified.The thesis focuses on the background subtraction method ViBe(Visual Background Extractor) and a particle filter tracking algorithm. The main work of the thesis is carried out to improve these two methods and as follows:1. In order to eliminate the effects of noise and light mutations and obtain accurate foreground objects, the thesis presents an improved moving target detection method that based on ViBe algorithm and associates with SILTP operator. At first, the improved method computes the SILTP value of each pixel, turns the background model into SILTP model and determines the foreground object by comparing the SILTP values. Then, the improved method’s background model update mechanism is the method of replacing abnormal pixels and uses multi frames to initialize the background model. Finally, the thesis uses some video streams to verify the effect of the improved algorithm and the result of experiments show that the improved algorithm has more robust to illumination changes and noise.2. A fast particle filter tracking algorithm based on a priori knowledge is proposed. The effect of particle filter is proportional to the number of particles, while the increase of particles will greatly accelerates the computational cost and reduces the speed of tracking. The proposed method uses ViBe algorithm to obtain the foreground objects and takes the foreground objects as the priori knowledge. At the phase of tracking, the proposed method increases the weight of foreground particles, sets the threshold of the weight, discards the invalid particles and reduces the frequency of resampling. By reducing the number of particles and increasing the weight of the effective particles, the speed of the particle filter is improved at the same time as the tracking performance is ensured. The result of comparative experiments shows that the proposed method performs better than the standard particle filter algorithm on speed.