Research On Image Multi-target Tracking And Tracks Keeping Based On Probability Hypothesis Density Filtering

The moving target detection and tracking in video images belongs to the front-end processing parts of the machine vision. For a long time, it is an ultimate goal to realize computer replace the human eyes to achieve target recognition and tracking functions by video image processing in the field of machine vision and computer vision. As the research direction of many researchers, choosing how to get useful information from video image sequence is becoming the theoretical foundation of the target detection and tracking. In real life, target detection and tracking in image have already had a lot of applications in many fields, which show the important research value in video images information processing. This topic concerned with the technology of multi-target tracking and tracks keeping, researching on multi-target tracking and tracks keeping in video images.Firstly, in the method of tracking by detection, image target detection is the premise of the track. Using aggregate channel features based on image color, gradient direction and gradient magnitude, combined with decision tree classifier based on boosting, generate the target detector to achieve accurate detection of images targets.Second, expounding the principle of target tracking based on Bayesian filtering framework in the top-down thinking. Research on Probability Hypothesis Density filtering recursive algorithm based on the random finite set approach, comparing the two different implementations of the simulation experiment, and finally confirmed using Gaussian Mixture PHD filtering algorithm for multi-target tracking.Third,combining with the characteristics of image targets appearance model and application conditions of Gaussian Mixture PHD algorithm, suitable for image multi-target tracking algorithm of Gaussian Mixture PHD, and verify the performance of the tracking algorithm by experimenting on pedestrian detection standard data sets.Finally, converting tracking problem into energy function optimization by building image multi-target energy function to solve the problem of filtering process can't achieve tracking trajectory and the target occlusion, merge and split etc. Carry on multiple iterations after processed by the tracking filtering algorithm, and finally realize pedestrian tracking trajectory of the video images.