| Video object detection and recognition is a core issue in computer vision research field, whose main task is to locate, detect, identify, analyze and judge the target in video image. The research purpose of the video target tracking is to make the machine obtain the ability of the human visual motion perception and the capability of recognizing moving targets in image sequences, thus providing important data for analysis and understanding of the target. Various disturbances appear in the acquisition process, such as blocking caused by a complex background image or by the target movement itself. Extensive research has been conducted and effective method has been proposed in the video object detection and recognition, but further efforts are still needed to develop a set of algorithm with better robustness for the new problems emerging in practical application.In this dissertation, the video single target tracking in complex environments has been the focus of research. In terms of target detection and recognition, target detection method based on template matching MCD distance and recognition algorithm based on AdaBoost are the key. In tracking, the application of window tracking algorithm is concentrated.The main research contributions of this study are as follows:1. This part introduces the target detection method based on template matching in detail, and several common target matching algorithm, and the MCD distance based template matching algorithm. The results of the MCD distance algorithm test show that, the template tracking method has the advantages of small calculation amount, matching accuracy etc..2. For object detection, this part proposes a multi resolution AdaBoost detect algorithm: first, use AdaBoost algorithm in a lower resolution to locate the target. Second, use AdaBoost algorithm in a higher resolution to get the accurate location.3. For object tracking, this dissertation proposes a method that is combined AdaBoost algorithm and window tracking algorithm. |
PDF Full Text Request