Research And Application On Detection And Tracking Of Object Based On Video Stream

In the various applications of computer vision, the object detection and tracking is one of the most important and basic tasks. Some of perspective applications include automatic driving system, robot control, video compressing, visual based control, motion recognition, human machine interface, medical imaging, augmented reality, and visual based intelligent surveillance system. Although the object detection and tracking has been studied for more than ten years in computer vision community, it is still an active research area. In present, there is not a kind of object detection and tracking system which is general, robust, accurate, efficient, and real time. Because the environment is complex, the scenes are cluttered, and there are a lot of problems in blocking and initialization. The video detection and tracking is one of the most difficult challenges among those tasks mentioned above.The highlights and main contributions of the dissertation include:In the aspect of object detection, in view of infrared target detection under the complex dynamic scene, a method was proposed for infrared movement target detection based on cross-entropy transition region extraction. Firstly, infrared image was carried on difference process by frame difference and background difference. Then binary image by segmentation algorithm based on cross-entropy transition region extraction. Finally, the complete infrared target was detected by the morphology filter.Shadow is one of some external factors reducing effects in the process of moving objects detection and tracking. This paper introduced the first order gradient algorithm of YUV model into HSI model in case that detection and tracking results were not ideal when moving target and background color were close. Some experiments proved this method can take great effect of detecting and eliminating shadow.In the aspect of object tracking,in the case of lost object in the traditional color histogram, a method was proposed for object real-time tracking basede on the weighted color probability. From Kalman filter object location and angle of direction, the weighted histogram predicted to quickly find the calculation of object tracking. To some extent, new methods to improve the measurement accuracy and stability, improve the tracking results. From the experiments, it can be confirmed that the new method can enhance the accuracy of video tracking process and have some practical value.