Method for detecting and locating emergent abnormal event of group

Abstract

The invention discloses a method for detecting and locating an emergent abnormal event of a group. The method comprises two steps: 1), group abnormal event detection; 2), abnormal event center real-time positioning. The method is advantageous in that (1), the method achieves the detection of common emergencies (panic, escape and gathering caused by fire, riot and fight) of public places (such as airport terminals), solves problems that a conventional detection algorithm is not apparent in behavior feature of the groups and is more sensitive to environment noise interference in a detection process, is high in detection rate for the above abnormal events, and is high in real-time performances; (2), the method solves problems that a conventional method is just used for the detection of abnormal events and is not used for the locating and subsequent processing of the event centers, can be suitable for various environments, can achieve the simultaneous detection of the positions of many danger resources in a scene, is high in accuracy, and is high in robustness.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a group sudden abnormal event detection and positioning method. Background technique [0002] The detection of group sudden abnormal events in public places is one of the hot issues in the field of computer vision research in recent years. With the development of computer vision, pattern recognition, data mining, artificial intelligence and other fields, the research on group abnormal behavior detection has also made great progress. Features are seriously interfered by real environment factors, noise interferes with crowd characteristics and affects the accuracy of subsequent detection results, and current methods often only focus on the detection and identification of abnormal group behavior in video image scenes, while ignoring the detection of dangerous events The positioning and tracking of the center. Due to the large crowd density and high concentration...

AI Technical Summary

Problems solved by technology

Ramin Mehran uses the social force model to reflect the interaction between the crowd in the scene, builds a local space-time cube and builds a bag of words model to detect abnormal behavior of the crowd; Cho classifies emergencies and builds a mixed event model to detect different types Abnormal behavior; Hu uses the energy model to describe the distribution of the crowd in the scene, so as to reflect the behavior characteristics of the crowd; R.Raghavendra uses the social force model, and uses the particle swarm optimization algorithm to minimize the interaction force to detect the abnormal behavior of the crowd. This method can Detect typical group abnormal behaviors, such as diffusion and directional movement. However, since the interaction force between particles is calculated for the entire image, this method is not ideal for detecting local abnormal behaviors; ChanM uses the HMM model to detect abnormal behaviors , this method needs to define an abnormal behavior sample in advance, and then use the predefined original space semantic expression to describe the defined abnormal behavior pattern, this method can avoid defining a large number of normal behavior models, but for some complex and unknown abnormal behaviors, the detection effect is restricted; Chen proposed to detect the abnormal behavior of the crowd by using the characteristics of crowd acceleration characteristics that produce mutations under abnormal circumstances. Method For emergencies where crowds move violently but the acceleration characteristics are not obvious, the detection results are easily disturbed; Duan uses a hybrid dynamic texture model to detect abnormal behavior, first estimates the parameters of the hybrid dynamic texture model, and then separately from the time domain The abnormal behavior is detected by the two factors of time domain and air domain, and the abnormal index of the abnormal behavior in the corresponding time domain and air domain is calculated. Finally, the time domain and air domain factors are considered comprehensively to judge the abnormal behavior. This method combines the time domain and air domain However, this method cannot accurately determine the specific location of the hazard, nor can it track the location of the hazard in real time. Variety

Although the above method can detect sudden abnormal behavior of the crowd in the scene and has made some progress, there are still problems such as environmental adaptability, scene applicability, and behavior type applicability for crowd feature extraction, and there is no such method in the current method. Locate and track the location of the event hazard source with more practical application value

Images