Study On The Video-based Moving Target Detection And Tracking Techniques

Video-based moving target detection and tracking is an importantand active research focus in the field of computer vision, imageprocessing and pattern recognition. It plays an indispensable role in awide range of applications involving smart video surveillance, industrialinspection, medical analysis and military. There are enormous businessopportunities and economic benefits. Academic and industrial circlesattach great importance to it. Therefore, real-time and robust researchon video-based moving target detection and tracking is significant to acertain extent.Traditional means of moving target detection cannot meet thedetected speed and precision at the same time because each mean hasits advantages and disadvantages. Therefore, optical flow method formoving target detection based on accelerated function is studied in thepaper for solving the problems of high complexity and low speed inconventional optical flow computation. Firstly, Harris corners areextracted from grabbed frame images of real-time video. Then cornerimages are generated on the basis of corner coordinates. Secondly, anefficient computational technique based on multi-grid is adopted tooptimize the optical flow estimation on the consecutive corner images.Therefore, real-time movement can be detected by optical flow vectorfield. And calculation is reduced because corner images andaccelerated multi-grid computing algorithm are adopted. After manyexperiments on videos with moving targets in different backgrounds andrates of frame grabbing, experimental results demonstrate the real-timecomputational speed. This approach is applied to detect moving targetsin real-time videos and the results are very encouraging. The approach aims at accelerating optical flow computation, also being robust to thedisturbance of complex background due to only adopting corner images.For disadvantages of moving target tracking algorithm in real-timeand accuracy, single moving target tracking method in simplebackground based on Kalman filter is studied in the paper. Firstly, oneframe without any moving target is grabbed from real-time video asbackground image. Then frame images with single moving targetgrabbed from real-time video are predicted and updated through Kalmanfilter. Thereby the locations of single moving target in video sequencesare obtained. Kalman filter using recursive filtering method describesdynamic system state sequences with the dynamic state equation andobservation equation. The method has a small amount of calculationand real-time computation. Finally, the single moving target trajectorycan be extracted and achieved through acquired location points bymorphological processing.Based on analyzing and studying the common video-based movingtarget detection and tracking system, experimental modules forvideo-based moving target detection and tracking are designed and built,which mainly involves computer, camera and frame grabber. With thehardware and data access method made up of Visual C++ programmingand image processing package, it is feasible to detect and track movingtarget for real-time video.