| Intelligent video analysis technology is an important technology in smart city development.Video-based abnormal event detection is an important application of intelligent video analysis technology,which makes great contribution to public security.This paper presents a viable solution for the crowd abnormal event detection,as described below:(1)Foreground extraction.This paper used Vibe based method to extract foreground in abnormal event video.To solve the slow elimination of ghost area,this paper update background model in steps.In the early of the algorithm,this paper cancel the probable update mechanism to accelerate update background model;in the late of the algorithm,this paper recovery probable update mechanism to inhibit the hollowing at target center in foreground.To solve the error detection of classic Vibe,this paper proposes using each pixel and its neighborhood foreground result calculate by classic Vibe to fix the final result.Experiments shows that our method can get the state-of-the-art performance.(2)Optical flow extraction.We extract the optical flow from foreground image.In order to highlight the contributions of the motion features to the detection of abnormal event,we proposed the concept of separation of optical flow velocity,which is filtered optical flow in low-speed part of the information,and remain the high-speed part of the information.Then use the morphological filtering eliminates noise in high-speed part of the optical flow.(3)Abnormal event detection.To be able to detect the occurrence of abnormal event,this paper propose a scoring mechanism for abnormal events by calculating the mean score and block score of high-speed part of the optical flow,then get the total score of abnormal event image.Based on this scoring mechanism,this paper using the classifier based on Gaussian mixture model to detect abnormal event.A large number of experiments show that our method can detect the occurrence of abnormal event effectively and promptly. |
PDF Full Text Request