Multi-Object Detection And Tracking Technology Research In The Dynamic Background

With the high-speed development of computer vision, video target tracking has become a hot issue in the field of computer vision, and under the background of dynamic target tracking more relative to the static monitoring or single target tracking is more complex, more challenging, so more people's attention and research. First of all,this problem is very wide range of aspects, involved in image processing, data processing, pattern recognition, signal processing, and other related fields. Secondly, the dynamic background of the multiple target tracking in the civilian security of intelligent city, traffic, aerial monitoring is of great significance, in military weapon guidance at the same time, the uav reconnaissance also plays an important role.Combining with the actual situation and the project requirements, for uav platforms and aerial video data collection, using multiple target tracking algorithm to complete to ground motion target detection and tracking, and focused on the movement of impassability complex scenarios of multi-objective situation, puts forward the improved algorithm, realize the stability of target tracking and motion track update the display.System divide into two big modules, detection and tracking target detection module adopts the principle of motion compensation, the two consecutive frames feature extraction and matching, combined with RANSAC algorithm to calculate the motion compensation of perspective transformation matrix parameters., in turn, the background of the target detection can be converted to static, finally, frame differential method with the improved graphics processing and information extraction to detect the moving targets and extract the movement information.Multiple target tracking module scores two child according to correlation and track management module, data correlation is each new testing information associated with previous track a moving target, track management is responsible for update of track, generated, or loss of target track is deleted, stable continuous tracking, finally the stability of the tracking data presented in the video.Completed under the background of dynamic target tracking, summarize the result of the experiment and analysis, then the target variable and object shelter complex situation improved algorithm is put forward, into the kalman filtering multi-targettracking to target block has a good adaptability, through the experiment analysis the feasibility and robustness of the algorithm.Finally verify and analyze the algorithm of common and complex situation the feasibility and advantages and disadvantages of multi-target tracking, and complete the whole simulation visual interface of the algorithm.