Detection And Tracking Method For Large Number Objects And Its Application In Real Time Video

In recent years, with constantly updated hardware and software technology, research of various image recognition systems has been greatly developed. Intelligent pattern recognition, machine vision, digital image processing, artificial intelligence, multi-disciplinary comprehensive video tracking system, gradually become the hot issues of the field of computer vision. Detection and tracking of moving objects are two crucial aspects of the intelligent video tracking system, and then become important research topics.In object detection part, current situation in the field of related research is firstly introduced. After analyzing the appearance-based object detection methods, we decided to use Haar-like feature to extract object feature. Secondly, the basic AdaBoost algorithm mathematical model and AdaBoost based face detection system principles are introduced. And in order to solve the deficiencies of AdaBoost based face detection system, we proposed a Feature Pruning based AdaBoost (FPAdaBoost) algorithm and a confirmation and skipping detection scheme (CSDS). FPAdaBoost cuts off features at a certain cutting coefficient according to the classification error in each iteration of training process, which effectively speeds up the learning and greatly reduces the computational cost compared with the traditional AdaBoost. And CSDS employs verification and confirmation scheme in the conventional scanning process, which effectively eliminates false positive detections. The performance of proposed detection method was tested in face detection and the results show that, compared with traditional Adaboost detection method, the training time of FPAdaBoost dramatically decreases without suffering a decline in classification capability in the training process, meanwhile the false positive detection is significantly reduced due to employing CSDS in the scanning process. Thus, the proposed detection method can make the training process more efficient and make the results more reliable.In object tracking part, we firstly introduce the current situation in the field of research. By comparing multiple object tracking methods, we ultimately choose the probability model as the framework for object tracking. A detailed description of mathematical model of particle filter and how it is used for object tracking is shown afterwards. Then the HSV color histogram based particle filter tracking algorithm is introduced. The algorithm employs HSV color histogram as likelihood model which is robust for illumination changes by decoupling the color and intensity information. Meanwhile, the calculation of HSV color histogram is simple which makes it very suitable for real-time object tracking. After the experiment, the results show that the HSV color histogram-based particle filter can track the object robustly.Finally, we combined cascade classifier trained by AdaBoost with HSV color histogram-based particle filter into object detection and tracking system. The cascade classifier scans the image sequence to detect the object, and the object area is then passed to the particle filter to establish reference evidence model for tracking. After the experiment, the results show that the system is stable to detect and track object despite the changes of illumination and scale.