The Research Of Multiple Target Tracking Algorithm Based On Video Sequence

With the continuous development of science and technology, the target trackingtechnology such as radar, thermal infrared, laser, video has been applied more andmore in military application. Gradually, the video monitoring platform technologywhich based on visual sensor is becoming more and more mature in terms of civilianvideo monitoring, and the new target tracking system based on video sequence is alsoinvolved in the monitoring ranks, so in the case of without human intervention, it canautomatically monitor and track the target, or alarming for where need peopledeliberately to observe, as some place that ban pedestrians in and out, and trajectorytracking to pedestrians, etc.Video target tracking technology is one of the most popular topic of visionresearch, which has been widely applied to license plate recognition, behavioranalysis, face recognition, fatigue driving, posture recognition, and many other fields.Video target tracking system can be divided into single target tracking andmulti-target tracking from the number of tracking targets. And multiple target tracking,which compared to the single objective due to target more than single goal, so at thetime of tracking target, there are many problems such as overlap and shelter betweenin targets, which can be solved hardly. And this article is mainly analysis to targettracking of pedestrian in video. Because of pedestrian movement speed is slow anduncertain direction in walking, so it is more difficult to trace. The pedestrian trackcannot be deal with easily as traffic flow tracking, therefore, it increased complexitywhile doing video processing, so make the video more real-time and accuracy oftarget tracking research become a big challenge.Due to the multiple target tracking only use a specific tracking technology, it isdifficult to track the multi-target stably in complex background, such as the conditionof the target block, size and morphology change, background interference. On thebase of algorithm study and actual test to the multiple target tracking, this articleextractive multiple target tracking algorithm research that aim at multiple targetmotion detection and feature extraction. The main work are as follows:It has good motion detection algorithm of the robustness. Because this article mainly research for multiple target tracking algorithm in specific environment thatwill not be changed, due to the influence of illumination, YUV space and yards of thismodeling method is adopted to complete, and because of the code of this algorithmwith high efficiency and overcome the previous background extraction in largeamounts of computation, thus it accelerate the operation speed greatly. And signalenergy analysis is used to filter out the interference of leaves and shadow in the actualoperation.The multiple target tracking algorithm that based on moving object detection andfeature extraction. On the basis of motion detection, we can extract the characteristicsof target, and according to the similarity of target feature to determine and track, andadding similar as Cam Shift and Mean Shift method when tracking, and researchingthe target tracking in terms of center of gravity moving, so it is good for the targetfusion and shelter, thus to realize the tracking of multiple target.