Research On Ground Moving Object Detection In Rotational Complicated Background

The technology of ground moving target detection based on photoelectric-detected video sequence is the key technology in the investigation, aiming and precision-guided weapon system. Especially with the violent rotating motion background and image nonlinear deformation, the slow moving target detection technology is the key technology to be solved urgently.First of all, the airborne photoelectric detecting video sequence is processed, including image sharpening, image noise filtering, image threshold segmentation, in order to provide basical images to realize the target detection algorithm. Then, this paper put forward a method to describe image complexity based on information entropy which reflects the image local information, as soon as possible to fully reflect the information of image details and complexity. And then, the evaluation of image quality after pretreatment based on the image clarity evaluation and image salt and pepper noise assessment is proposed, in order to judge whether the image quality was improved and provide better quality image for background modeling.Aiming at the problems of nonlinear deformation of image in the complex background, a method of slow target detection based on nonlinear model is proposed. Firstly, the law of background motion is studied and the center position of background rotation region and the direction of background rotation is determined. After that, we apply the nonlinear transform to the background modeling of rotational motion: we detect the corner points of image by Harris, obtain the optical flow of feature points through pyramidal Lucas Kanade optical flow method, filter the optical flow of feature points, obtain precise matching feature point set and estimate nonlinear model parameters by least squares estimation. Then we compensate background by means of bilinear interpolation and evaluate the effect of image compensation. The evaluation index of image background compensation is proposed, and the effect of image compensation is evaluated. The images difference is obtained by the background modeling,and the iterative algorithm for segmentation threshold based on difference image is proposed. Due to the background of continuous video images have a certain correlation, the segmentation thresholds of sequence difference images are analyzed, and the necessary correction of the difference image segmentation thresholds is done. We Mark the target with the box to accurately detect the slow moving target in the rotational complicated background. In order to avoid the holes of target detection based on difference method, the algorithm of edge detection based on region growing rule is proposed for completely detecting the moving target.