Research On Optical Flow Technique And Its Application In Moving Object Detecting And Tracking

Concerning the application destination of Moving Object Detecting andTracking(MODT), a significant technique——Optical Flow Technique(OFT) has beenresearched and used to estimate two dimensional movement in this dissertation. This paper also analyzes the application status of OFT in MODT. Finally, basing upon the method of optical flow equal sampling, a MODT system has been realized. All these methods and algorithms referred in the paper have been validated by means of simulation.The derivative optical flow(DOF) technique has been studied first. A purposeful solution is proposed and realized in order to solve the problems of DOF in respect of gray-discontinuity and noise-sensitivity. The filter mode, differential operator, smoothing-restricting method as well as multi-distinguishing strategy has been improved in our model. Satisfying results has been shown in our practical experiment.Subsequently, the feature optical flow(FOF) has been concerned in this dissertation. Point matching is the substance of FOR Both point matching using exhausting-search and point matching using optimum-estimate have been analyzed. The former has higher precision but larger calculation amount while the latter has the reverse state. To balance the precision and the amount of calculation, an improved method of point matching is proposed. The method enormously speeds up the matching and keeps up higher precision by searching locally under the guide of optimum-estimate.In the end, we have investigated the application status of OFT in MODT. In face of the contradictive state: the abundant information quantity but larger calculation amount of DOF and the deficient information quantity but less calculation amount of FOF, a MODT method which combines DOF and FOF is analyzed and an OF equal sampling based MODT method is proposed. The simulation experiment with 700 points sampled in each frame of several image sequences indicated that the processing time between frames is less than 400 ms even on a normal PC environment. The problem between information quantity and calculation amount has been solved elementarily.