Research On Multi - Target Tracking Technology Based On Multi - Camera

With the gradually development of information technology, the technology of video monitoring is improving steadily and gains its significance on parking areas, airports, supermarkets and other public places except for traffic fields. However, in modern society, the monitoring scene is becoming more and more complicated, which causes that the single-camera video monitoring system cannot meet the requirement of human society, therefore multi-camera video monitoring technique has become the major focus of researchers. Moreover, the issue of multi-target tracking in multi-camera is a priority for all.This paper generally focused on the problem of multi-target tracking in multi-camera video monitoring system, it can be divided into two parts:one is the data association problem of multi-target tracking in a single camera, the other is issue of track-to-track correlation between multi-camera. For the reason that traditional multiple target data association algorithm has high error rates and low real-time performance, we propose a new data-association algorithm based on multiple features, in which the incidence matrix of traditional multiple hypotheses tracking (MHT) data association is combined with the moving features, HSV color features, LBP features and optical flow. The experiments indicate that the association probability and real-time performance of the algorithm is improved significantly. Moreover considering the track association, the calculation amount of tracking association algorithms increases at an exponential rate to the growing number of cameras. As for this, the bionic ant-colony algorithm is used to calculate correlation matrix, which is proved that the algorithm can reduce the association time when the number of camera is three by the experiment. Furthermore according to the shortcomings of the algorithm, which has a long cycle and low convergence speed, we put forward an improved ant colony algorithm. Inspired by genetic algorithm, this algorithm improves the searching ability by crossing the track of one track-pair. The experiment confirms that the algorithm can enhance convergence speed to a certain extent, and has a higher precision of association. Consequently these two methods are combined into the multi-target tracking, and it has been verified that these two kinds of algorithms are feasible on experiment.