| The security detection of intelligent kindergarten based on video surveillance refers to the real-time and automatic detection of the young children's abnormal behavior in intelligent kindergarten with the video surveillance as the carrier,so as to ensure the security of the children.On the one hand,today's society is in the information age,the existence of Internet of Things allows us to obtain the massive data;on the other hand,with the improvement of people's living standards,parents pay more attention to children's growth and health in physical and mental.Compared with the traditional kindergarten,the security detection of the intelligent kindergarten also needs to be more perfect.Therefore,this paper is committed to the research on the key technologies of intelligent kindergarten security detection.We regard the video surveillance as the carrier,accomplish the research on the security detection of intelligent kindergarten.According to the complexity of the video surveillance scene and the requirements of abnormal detection,the security detection of intelligent kindergarten can be divided into simple scene detection and complex scene detection.Firstly,we propose three kinds of rule-based abnormal behavior detection algorithms,which are cross-border detection,rule-based detection in closed area and fall detection.They are mainly used for the abnormal behavior detection of young children in simple scene.These three kinds of detection algorithms are simple to implement,and can achieve good detection results with the detection rate of 100%,100%and 92%.While they need to be specified the abnormal rules in advance,so the scalability of detection algorithms are not strong.Secondly,we propose an improved algorithm for moving multi-target tracking based on Camshift algorithm and Kalman algorithm.And further,an algorithm for detecting the individual motion anomalies of the young children based on the motion model is proposed according to the tracking algorithm.Among them,the proposed multi-target tracking algorithm is used to realize the function of multi-target tracking,automatic initialization of new tracking,automatic treatment of failure tracking and so on.The proposed detection algorithm of individual motion anomalies establish a motion model with the smallest sub-blocks of the video surveillance image based on moving target information by using the proposed multi-target tracking algorithm,so as to detect the unknown moving targets.And this proposed detection algorithm is used to the real-time detection of the position,movement direction,movement rate and dwell time of the children in video surveillance.It has high accuracy and superiority,especially for the individual motion anomalies in scenes with fewer moving targets.The average recall rate is 92.6%and the average detection speed is 33.8 ms/frame,which is better than the other similar anomaly detection algorithms.Third,we propose a detection algorithm based on background difference algorithm and optical flow algorithm in terms of the motion characteristics of crowd state mutation events.It is used to detect the anomalies of crowd state mutation events when children gather together in the kindergarten.This proposed algorithm combines the advantages of the data processing speed with background difference algorithm and the result accuracy with optical flow algorithm.The AUC of UMN video database can reach 0.96,and the average detection speed is 25.8 ms/frame.Compared with the other similar detection algorithms,it has a great improvement in comprehensive performance.Finally,the crowd density estimation method based on pixel statistics is used to distinguish and combine these various security detection algorithms proposed above to accomplish a complete security detection algorithm suitable for intelligent kindergarten.Furthermore,we achieve the key technologies of intelligent kindergarten security detection based on video surveillance on the platform of Microsoft Visual Studio 2010 with C++ language by using OpenCV. |
PDF Full Text Request