A Method Research Of Moving Target Detection And Tracking Based On Omni-direction Vision

Omni-directional Vision and moving target detection tracking is the main research focusof the field of computer vision; the combination of two technologies has broad applicationprospects. The characteristics of the omnidirectional vision sensor is able to obtain thehigh-resolution images of the scene within the scope of horizontal direction360°and thevertical direction240°, and which can solve the deficiencies of the traditional vision sensorscan only be observing the local information. The article use Omni-directional Vision system,The commencement of the panoramic image on the visual sensor acquisition of imagesequences, And the expanded image sequences of moving target detection and tracking.The images collected by Omni-directional Vision sensors do not meet the observationhabit of human, so it should be expand. In order to improve the algorithm speed, theprocessing of expanding adopts the adjacent difference algorithm to optimize, and describesthe software of image expand. And describes the image expand Software.First introduces some common methods of target detection, and study of the advantagesor disadvantages of each method in-depth, then proposed improvement algorithm. At thesame time, In order to meet moving target detection based Omni-directional Vision forreal-time requirements, we propose a target detection algorithm based on adaptive Gaussianmixture model. This method has a good effect of target detection in complex environments.Through introducing the initialization of Gaussian mixture model parameters, selecting theGaussian number, real-time updating of background, to understand the principle of Gaussianmixture modeling. Understand the principle of Gaussian mixture modeling. And then throughthe experimental comparison, adopt the statistical average background modeling method andthe Gaussian mixture background modeling method to detect target in which the expandsequence of the full range of visual images of complex scenes, and to verify the superiority ofthe algorithm.This part uses the Mean Shift algorithm to track and process moving targets based on thefull range of visual and focuses on the principle of the Mean Shift algorithm and theapplication of target tracking. Mean Shift algorithm does not have predictive capability, poorrobustness, easy to lose the target in the complex video environment, so through a combination of Kalman Filter and Mean Shift algorithm to solve the above problem. In theoriginal Mean Shift algorithm in target tracking, the characteristics of the target selectionbased on the RGB color space, and this paper changed to the selected target feature valuebased on the HSV color space in order to improve the operation speed of the trackingalgorithm. Finally, through comparative experiment between in the case of complex videoenvironments and tracking targets suddenly accelerated to verify the correctness of themethod.