Research On The Detection Algorithm Of Abnormal Crowd State

Safety management in prison, people are generally concerned about whether the abnormal behavior of prisoners will bring harm or not. It is known to all that the traditional human supervisory, alarm system and access control are clearly limited. However, the abnormal behavior recognition based on video surveillance exerts a great power on reducing the tasks and pressure of the prison staff. Nevertheless, video image capture is only a basis of the abnormal behavior recognition of video surveillance system. Furthermore, the image sharpness or other issues determine the effect of the abnormal behavior recognition, but the image capture clarity is restricted in some situations, such as fog rain, dark night, etc. It is difficult to separate the different faces from those who wear the same clothes within the limitation of the present image processing technology. Therefore, based on ZigBee location technique, the abnormal behavior recognition applied to prison security management will be important and practical. The thesis was started to study the above issue, and the main research parts are as follows.1, Two methods to extract path, video and wireless, were analyzed and compared, as well as the current common location technique, algorithm and the dominating abnormal behavior recognition were studied in the thesis. The thesis deeply studied the merits and demerits of the current location technique, algorithm and the anomaly detection method, thus selecting ZigBee wireless communication technology as the new location technique, and having a new abnormal behavior recognition method based on both semi-supervised learning and track.2, Three-square location algorithm was proposed. Aiming at complex situations and inaccurate location problem of trilateration in practical application, the author put forward a new three-square location algorithm on the basis of it and centroid method. The three-square location method has much higher practical value, because it is easier on the thinking and realization, and doesn’t need to consider various situations. Even if there are only two fixed node, the three-square location method can be used in larger error. Artificial experiment results also reflect that the method is much more accurate than the improved trilateration and centroid method.3, The abnormal behavior recognition process was analyzed and verified. The author applied Hausdorff to calculate the distance between the two tracks, employing spectral clustering and K-means as clustering algorithm, and picking the closest track from clustering central distance to characterize the central track of the average standard of this kind of track. During the process, deviation obeyed to normal distribution, deviation model was found, and the track was segmented. At last, according to the confidence intervals of the deviation to distinguish whether the behavior is normal or not.4,The abnormal behavior recognition system based on ZigBee was designed and founded. In accordance with the thinking of facing targets, the system contains embedded communication and PC monitoring platform which is clarified of five modules(serial port communication, location, database, abnormal recognition and database processing), and then the corresponding code was edited. Under the simulative scenes, the author committed a test on the abnormal behavior recognition which completed all the functions that the system has.