| With the development of economy and the progress of society, we witness the coming of information era. In the social and public environment, the staff tends to be intensive, the place tends to be complicated, the information tends to be massive and the event tends to be complex. It is higher required of the efficiency of processing method for these trends. Specially, the processing method is getting informatized and intelligentized in security and safety preservation. As is known that, the medium to transmit and restore information includes text signal, audio signal, image signal, video signal, etc., in which the visual signal including image and video, contains large amount of information. Therefore, with the promotion of technical demand, intelligent video analysis based on surveillance camera system has been greatly developed, from single camera to multi-camera. It manages to preserve security and surveillant, and control risk and disaster, based on efficient behavior recognition and event detection method. Owing to the technical support of it, the economic interests, social order and personal property safety has been strongly protected.In recent years, behavior recognition and event detection based on multi-camera system has been investigated deeply and applied widely. The technology of behavior recognition and event detection is getting complete, where the basic framework including feature extraction, object detection and behavior representation, classification or decision of behavior and event, has been widely verified and applied. Meanwhile, multi-camera analysis method has been developed based on single camera behavior recognition methodology to meet the complicated multi-camera environment. Among these tasks, the behavior representation modeling of object and event detection modeling across multi-camera are critical for behavior recognition and event detection in multi-camera system, which will impact the performance to a large degree.Among current technologies, there are many methods such as trajectory-based, region-based and so on to model object behavior representation, and there are many methods such as parametric models, graphical models, syntactic approaches, knowledge-based approaches, logic-based approaches and etc. These well-performed models and approaches would have big advantage in many special situations and under certain conditions, however, also have some drawbacks and limit in some cases. Some of these approaches may be sensitive to noise of object tracking and much dependent on accuracy of it, some may require complicated learning and configuring process, and some may rely on big amount of training data. This paper manage to integrate the advantage of some existing approaches, and overcome the drawbacks of them. We investigate the problem of object behavior representation and event detection model and propose a network modeling approach.In this paper, we propose a network-transmission-based algorithm to detect event of abnormal behavior in multi-camera surveillance applications. The proposed algorithm models the entire field of view (FOV) covered by the multi-camera system as a network. In this network, each node corresponds to a segmentation of the entire FOV and each weighted edge represents the behavior correlation between the corresponding segmentations. Based on this network, the proposed algorithm further models human activities as the signal transmission process in the network. Thus, abnormal activities can be detected by tow criteria of’network transmission energy’. Compared with the previous methods, the proposed algorithm is more general and is flexible to handle various multi-camera scenarios and configurations. Besides, it’s not necessary for the proposed algorithm to be trained under a great quantity of training data. A large amount of experiments have been conduct to verify the proposed algorithm and compare with other well-performed approaches. Experimental results demonstrate the effectiveness of the proposed algorithm. |
PDF Full Text Request