Based On O2O Learning Multiple Target Detection And Tracking Technology Research

In the field of video surveillance, Multiple target tracking technology determine many video target position, size, and its complete trajectory by computer. Video information include information in frame and information between frames. Based on motion information existing detection tracking system, only using motion information between frames, ignores the apparent frame object information, so the system can’t detect stationary state, and that target can’t be tracked from movement to stop into the background.For the problem that moving target detection system can’t extract the stationary state, use O2 O comprehensive detection algorithm, Online based on motion information between frames and Offline based on frame apparent information; Motion detection by information between frames extract motion targets; Apparent detection make up the lack of motion detection, extract stationary state targets; Learning both the extraction results, output comprehensive detection results.For the problem that motion detection detect targets which include some of the tracking target noise, offline train random forest classifier of object, use classifier, to filter out the noise and improve the tracking accuracy, reduce the trace amount of calculation.For the problem that target cannot be tracked from movement to stop into the background, use multiple target tracking method based on the fusion of feature matching and data correlation, data correlation methods to solve multi-objective related problems between frames, feature matching method to judge whether target disappear.This paper main content is as follows: First, research motion detection based on background difference method to obtain target coordinates, size, speed and other state data. Research apparent detection based on O2 O learning random forest classifier to obtain the target coordinates, size, characteristics of categories such as state data. Then, use forward multiple target tracking system based on fusion of feature matching and data association to get the merged data and output target state. Finally, use forward a multi-target detection and tracking system based on O2 O motion and apparent. When the tracking system initialization, use motion information online and apparent characteristics classifier offline to detect all targets, then use feature matching and data association to track multiple target.