Applied Research On Target Tracking System Of Multiple Cameras

Moving target tracking based on video monitoring technology belongs to a branch of artificial intelligence, involved in pattern recognition, image processing, computer vision and machine learning disciplines.Single camera target detection and tracking has been relatively mature, made more research achievements, but there are still many problems need to be solved.Such as the target block, similar background interference, the dynamic background tracking, etc.Multiple cameras target tracking is a research hotspot in recent years, transition goal is key technologies to realize target matching and scene switching tracking, and is the problem that the multiple cameras monitoring must solve.This topic in the research on single camera motion target detection and tracking algorithm based on the new method is proposed, and focuses on the dual camera view overlapping case view line generation and target handover algorithm, the main work is as follows:1. Real-time background update target detection algorithm is adopted.Firstly the initial background image is generated by gaussian mixture background modeling;Then by combining three frame difference method and background subtraction method to detect moving targets;Finally, the target area and the motion area respectively take different background update policy.2. Using particle filter for the improved Camshift target tracking algorithm, combined with Camshift algorithm to improve the particle filter algorithm.At the same time solve the particle filter that needs a large number of the particles, and particles’ slow convergence speed and tracking real-time performance problems and shade Camshift algorithm of target tracking robustness low problem.3. This paper proposes a judging and projective invariants algorithm, based on a rapid SURF feature points matching methods.First SURF algorithm quickly get two adjacent video camera with a background image of the overlapping area of feature points, then use RANSAC algorithm optimize the feature point vector, removal of error matching points.Iteration based on RANSAC algorithm, selecting four of the best match point, finally, according to the boundary point two background images and projective invariants for overlap in the two images, and calibrate, generate sight line.4. This paper proposes a vision based on boundary geometry target handover algorithm of coordinate transformation curve matching method.First of all, if there is a moving target real-time detection public eye area, and then to carry on the coordinate transformation, and under the new coordinate system, statistics of target pixels, in the end, will get the pixel statistics curve matching and calibration.