No abstract

An efficient method is introduced for detecting humans in surveillance video. The method improves the performance of multiscale human detection through the use of a scalemap, and does not require knowledge of the camera parameters or the use of additional devices. A scalemap is a map that links each position in the observed image to the optimal detection scale. The proposed method efficiently reduces the computational costs by estimating the scale of interest and the region of interest based on the scalemap, while maintaining the accuracy of the detection. It is experimentally shown through an experiment that the proposed method can improve both the accuracy and the efficiency of real-world surveillance videos.

A motion pattern analysis method for abnormal movement detection is introduced. It analyses motion patterns and detects abnormal movement using partial trajectories, which can be obtained in crowded scenes and are more effective than local motion. In addition, the proposed method is able to deal with noisy data, which is a major cause of false alarms. The experimental results from real-world traffic scene datasets show that the proposed method improves on previous, local motion-based methods.

No abstract