Crowd abnormal event detection method, device, system and storage medium

Abstract

The invention provides a crowd abnormal event detection method, device, system and a storage medium. The method comprises the following steps: obtaining a video to be detected, and extracting pedestrian skeleton posture according to part or all video frames of the video to obtain pedestrian skeleton information of each frame in the video frame; For each pedestrian in the video frame, the skeletoninformation time series of the pedestrian is generated based on all the pedestrian skeleton information of the same pedestrian in the video frame. The skeleton information time series of two or more different pedestrians in a video frame are combined to form a sequence group, and the sequence group is classified by using a cyclic neural network to detect whether an abnormal event occurs between the pedestrians corresponding to the sequence group. The crowd abnormal event detection method of the invention obtains the skeleton information of the pedestrian through the skeleton posture extractionalgorithm of the pedestrian, and then discriminates and detects the abnormal situation based on the skeleton information of the pedestrian time sequence by adopting the circulating neural network, sothat the accuracy rate of the crowd abnormal event detection can be greatly improved.

Description

technical field [0001] The present invention relates to the technical field of pedestrian detection, and more particularly to a method, device, system and storage medium for detecting abnormal crowd events. Background technique [0002] Among the many applications of video structuring, pedestrian behavior recognition and abnormal event detection are very important, and can play an important role in urban security monitoring tasks. [0003] At present, for the detection of abnormal crowd events (such as fights) in videos, most of the methods are to extract the motion vector information in the video, obtain the descriptive features of the violent movement through the motion vector information, or directly train the classification on these motion information model to get prediction results. However, this method does not understand the behavior of the characters in the video in essence, so it may appear that non-abnormal events (such as square dancing) are identified and detect...

AI Technical Summary

Problems solved by technology

However, this method does not understand the behavior of the characters in the video in essence, so it may appear that non-abnormal events (such as square dancing) are identified and detected as abnormal events, which makes it difficult to detect crowd abnormal events. high accuracy

Images