| The monitoring systems are very common in our daily life, but few of them are real intelligent. Especially when they are used to monitor suspicious persons, the field of view is very large and the target is far from the camera, which may cause the problem that the conventional pedestrian detection algorithms are not likely to detect the suspicious persons correctly, much less to identify suspicious persons automatically. In this paper, we focus on the method to detect the small pedestrian correctly and then track the target automatically. The main contributions of this paper are summarized as follows:Firstly, by studying the PTZ control methods proposed in the literatures, we find that when the field of view is very large, few methods can give an accurate control result. To address this problem, we propose an accuracy control method based on PTZ system. Experiments on some common monitor-scenarios demonstrate that the proposed method is outperforms the state-of-the-art control methods in terms of the accuracy.Furthermore, to solve the problem of a high miss-detection rate caused by using the HOG algorithm alone to detect the pedestrian, a coarse-to-fine pedestrian detection method is proposed, which can accurately detect small pedestrians in a large field.Particularly, utilizing the information of the size, the movement, the proportion and the spatio-temporal property of the pedestrian, we propose a small pedestrian detection algorithm based on the adaptive confidence Gaussian Mixture Model(GMM). By doing a series of operations such as detecting the moving target, generating the target’ trajectory, calculating the confidence of trajectory, the proposed method can correctly detect the small pedestrians in the video stream.Besides, an active pedestrian detection and tracking system using a PTZ camera is designed and implemented in our paper, which can be used to monitor the pedestrian in a large scene, and the monitor area can reach up to 5,000 square meters. |
PDF Full Text Request