Research On Moving Object Detection And Tracking With Auxiliary Road Information

Moving object Detection and tracking is a very important branch of the video processingtechnology, which is widely used in intelligent transportation, biomedical, industrial production,public security and military applications. UAV platform which is equipped with video sensors todetect and track ground moving objects in real time can help the latest battlefield intelligenceinformation to be got, which is of great significance in the battlefield reconnaissance andsurveillance missions. However UAV image sequence background is complex, and the movingobjects in the sequence has small size but large displacement. Traditional methods of detectionand tracking objects are difficult to detect the moving objects accurately and track thempersistently. The research purpose of this paper is to bring road information into the field ofdetection and tracking moving objects in the video from UAV based on optical flow analysis toachieve detection and tracking small objects. The major work and achieved results now has beencompleted in four aspects as follows:1. The basic principles and calculation methods of optical flow technology are describedsystematically. Optical flow algorithm based on feature matching and multi-resolutioncoarse-to-fine hierarchical strategy which can be used for large displacement motion estimationare briefly analyzed. Comparison experiment is taken to lay the foundation of a new largedisplacement optical flow algorithm which is to be proposed.2. A new large displacement optical flow algorithm which combines feature matching andderivative gradient is proposed to achieve motion estimation of small objects with largedisplacement in UAV video. Using the constraint of correspondence feature got from featurematching with traditional optical flow model, large displacement optical flow model is built andthen large displacement optical estimation is achieved, combining multi-resolutioncoarse-to-fine optimization strategy.3. With road information, roads buffer is built basing on the principle of buffer analysis,using road node coordinates as the basis of geographic data. The idea of sub-blocks to determinethe road detection area is used on the basis of the road buffer, which effectively reduce thedetection area. Then Gaussian mixed probability model and EM clustering algorithm is used toachieve the fast detection of small object with large displacement in UAV video.4. The basic principle of Kalman filter is analyzed. Combining with the constraint ofposition and direction of the vehicles by road information, the Kalman filter model under roadconstraint is proposed after taking the constraint into Kalman filter. With predicting the state ofvehicles by template matching based on area-based matching, tracking the cars in UAV video under road information persistently can be achieved.