Research On Ground Moving Object Detection And Tracking Method In Rotational Background

Typical moving ground object detection and tracking based on photoelectric-detected videoimage is the key technology for aircraft investigation aiming and precise strike. When aircraft makes asharp turn, which result in the rotational movement of background in the images, correctly detectionand tracking of ground target in rotational background is one of the urgent key issues in thistechnology.Basing on an intensive study on airborne photoelectric-detected imaging system, a model isconstructed to represent the motion of background pixels from one frame to the next. By researchingthe motional pattern of both background pixels and target pixels of airborne photoelectric-detectedimages, a method, which is based on motion vector distribution ratio between background and target,is proposed to evaluate the level of ground background rotation. And according to this method,rotational backgrounds are divided into2classes: gently rotational background and severely rotationalbackground.For the situation of moving object detection in gently rotational background, the image featurepoint sparse optical flow field is calculated by using Shi-Tomasi Harris feature point detectionalgorithm and improved color pyramid Lucas optical flow algorithm. Basing on the sparse opticalflow field of image feature points, we estimate the center motion vector of the background from,classifies image feature points, divides input image into regions and finds out the region of movingobject with the help of target confidence to perform the moving object detection algorithm. Forseverely rotational background, moving object detection algorithm is also based on the sparse opticalflow field of image feature points to estimate the parameters of perspective background motion model,reconstruct the background motion vector field, compensates background motion for the images, andby differencing binary image and morphologically processing the binary difference image to achievemoving target.Affected by various factors, the status of moving ground target is complex in airbornephotoelectric-detected video images. Single tracking algorithm can hardly achieve great performancein all statuses. For3kinds of moving ground target statuses: deformation, occlusion and high-speed,3continuous predict-search-track modes are proposed to deal with corresponded status. In order toimprove the stability, a tracking quality coefficient is also proposed to intelligently judge target statusand determine whether updating the target template or switching to the best match tracking mode for accurately moving ground object tracking.