Pedestrian Counting Algorithm Research And Achivement Based On Video Sequence

Pedestrian detecting and tracking as a hot and difficult problem in the computer vision, is studied by many scholars and research institutions as an important topic. The application scope of the subject has been widely promoted, such as intelligent monitoring, human body abnormal behavior monitoring and pedestrian counting, which are all based on this object. Among them, pedestrian counting has important research value in practice, this paper mainly conduct a study of pedestrian counting.The human targets, as a special non rigid moving object, has its own difficulties in targets detection, such as its non rigid characteristic, and the complexity of the scene, and the problem that all targets detection will face on and so on, such as the changes in illumination and the target of each other the problem of occlusion, because of these problems, all these problems have decided that the traditional background subtraction method can not meet the requirements in terms of accuracy and precision. Yet the histogram of oriented gradients method, which is called HOG feature extraction method in the article with high accuracy when it is referred, is hard to meet the real-time requirement in practical application because of its complex calculation. So, based on the above two problems, the paper proposed an optimization method, combing the mixed Gauss background modeling method with the histograms of oriented gradients method. The improved method solves the low accuracy problem that only use of mixed Gauss background modeling method, at the same time, it also improves the detecting speed of HOG feature extraction. Through the experiment verification, it is proved that the improvement algorithm has obtained the certain effect.In the pedestrian counting process, in addition to the need for pedestrian detection, the tracking problems after detecting also need to be solved, which is another difficulty. This paper first introduced the common tracking algorithm, in which the principle of Camshift, its advantages and disadvantages are mainly introduced. Furthermore, based on this algorithm, the improved algorithm is put forward that is suitable for this scene. By studying, we found that the main disadvantages of this algorithm is that during its initialization, the tracking target need to be manually selected, what’s more, multiple targets occlusion seriously prone to target lost phenomenon. The improved algorithm in this paper is mainly to do optimization in the two aspects. Firstly, the tracking in the paper is based on target detection, so the initial target is marked target in detection progress, and against the lost phenomenon, the thesis proposes adaptively search window function parameters, this method can adaptively adjust the search window and inhibition the noise that will affect the tracking. Specific approach is:when the object disappears, firstly, forecast the motion trails of the target, when the target reappears, the algorithm will be able to complete the target tracking.At last, based on the above algorithm research, the paper designs a pedestrian counting system that based on Visual Studio2008and OpenCV2.0, and realizes the system. It contains the pedestrian detecting and tracking counting function. Finally, the thesis tests the system by three sample videos, which proved the availability of this system, besides that, the paper also analyzes the faultiness of system, and at the same time, points out the research direction in the future. The innovation of this paper is mainly aimed at optimizing the common algorithm in the two respects pedestrian detection and tracking, and realizes the pedestrian counting system that based on the proposed theory.