Abstract
Camera sensor networks have developed as a new technology for the wide-area video surveillance. In view of the limited power and computational capability of the camera nodes, the paper presents an abnormal behavior detection approach which is convenient and available for camera sensor networks. Trajectory analysis and anomaly modeling are carried out by single-node processing, whereas anomaly detection is performed by multinode voting. The main contributions of the proposed method are summarized as follows. First, target trajectories are reconstructed and represented as symbol sequences. Second, the sequences are taken into account using Markov model for building the transition probability matrix which can be used to automatically analyze abnormal behavior. Third, the final decision of anomaly detection is made through the majority voting of local results of individual camera nodes. Experimental results show that the proposed method can effectively estimate typical abnormal behaviors in real scenes.
Highlights
Camera sensor networks consist of low-power microcamera nodes, which integrate the image sensor, embedded processor, and wireless transceiver
The anomaly detection using trajectory analysis is divided into three phases
The sequence of Euclidean distance of target trajectory extracted from the raw image is trained and modeled in order to obtain the anomaly detection threshold
Summary
Camera sensor networks consist of low-power microcamera nodes, which integrate the image sensor, embedded processor, and wireless transceiver. The existing works mainly focus on maximizing the detection accuracy and detect abnormal behaviors through a single visual camera without involving the information fusion and interaction of camera nodes These methods are inapplicable to sensor networks because of the node’s limited power, memory, and computational capabilities. The work in [10] proposed a HMM-based approach for detecting crowd behavior by using a heterogeneous sensors network comprising visual cameras and a thermal infrared camera. The two approaches merely use the features extracted from heterogeneous camera sensors to model normal behavior, which is executed by a centralized supervised learning method. It is infeasible in distributed camera sensor networks.
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call