Vision-based Moving Object Detection And Tracking For Unmanned Helicopter

Unmanned helicopter which has the advantages of small size, light weight, low cost,convenient hiding,flexibility, is widely used in military and civil areas. With the development of computer vision technology, vision-based moving object detection and tracking become one of the core technologies in UAV applications. Currently, object detection and tracking for static scenes technology has been matured. Because the relative motion between background and moving object in the graphics sequence captured by sensor is made by UAV platform's moving, object detection and tracking technology become more complex. This paper aimds at the detection and tracking algorithm for moving object for unmanned helicopter, and studies on some algorithms related to motion compensation, object detection and tracking. The main works are as follows:Due to the relative motion of the background in graphics sequence and moving objects can not been detected directly from the graphics sequence, the studying on background motion compensation is deeply discussed. An new improved feature points detection method based on SUSAN is presented. This algorithm improves the original 37 pixels circular template which is instead of the circular template with 17 pixels and using an adaptive threshold method. The amount of computation is reduced and the robustness of detection method is improved with this method. The feature points are cstracted successfully with this method. The feature points are matched by the method based on adaptive cross searching and region matching. False matching points removed by RASUAC algorithm and background global motion parameters are acquired by establishing the parameters model. Global moving background are compensated and the background become sraticizing.The method of moving objects detection is described. To the characteristics of the UAV moving objects, an improved moving object detection algorithm based on improved three-frame difference is proposed. Using information of three frames and seting the threshold adaptively, The algorithm improves the robustness and accuracy. Combining morphological filtering, moving object is detected and located accurately.With good versatility and robustness, the particle filter is widely applied in object tracking. Since the UAV scene background is complex, single feature can not be descrip object accurately. This paper chooses color feature and edge feature as characteristics of object and integrates these two features dynamically, so moving object is tracked accurately with the method of particle filter. This approach adaptes the changes of the background and object in UAV graphics sequence effectively.