Implementation Of Mining, Monitoring And Identifying System For Tourism Emergencies

Emergency is regarded to people’s safety and benefit, besides, the mining, monitoring and recognizing of emergency is a popular subject and plays a very important role in a lot of aspects. With the collection,mining and analysis of related scenic information, using the scenic spot video to detect the crowd number and abnormal event of scenic spots is an important goal of current tourism activity monitoring research. With the development of the national tourism industry in recent years, tourists pay more attention to their benefit and safety during their travel. The re-search of tourism emergency monitoring and identification based on sur-veillance video has very important realistic significance.This thesis focuses on video-based monitoring and identification of tourism emergencies, and also develops and implements the relevant verification system. The main work is as follows:(1) The method of crowd estimation based on SVR regression is pro-posed, which uses GMM to extract the foreground and calculate the orig-inal picture and the foreground to get the mask of foreground to model the background of video. Then, surf algorithm is used to extract features of foreground so that the feature points can be clustered by clustering al-gorithm. Camera height, angle, and density of feature points and cluster-ing information are combined to establish the local crowd training set based on PETS2010 data set. Then a model between the number of scenes and the feature points is established. The SVR algorithm is used to predict the number of people.(2) A method of crowd abnormal monitoring based on crowd density information distribution is proposed. Through the crowd clustering and SVR regression algorithm, the local density of the crowd information is obtained, reducing background noise of the traditional social force model.At the same time, the particle mass parameters, aggregation parameters and local density are correlated to calculate the crowd pressure. In the abnormal monitoring, the anomalous feature is often associated with the temporal and spatial characteristics of the scene. The method of mul-ti-frame superposition is used to correlate the abnormal characteristics with the temporal and spatial characteristics of the pressure.(3) A method to monitor the crowd abnormally with the VIF feature and social force model is proposed. The crowd distribution is low in out-side area like square, while the threshold value is also not high. However,in place like museum, the crowd distribution is very high, as well as its society force threshold value. Meanwhile, the visitors flow rate varies during different time even in same location. The crowd reaches the peak on holidays, while social force increases in the same time. In the simple abnormal monitoring algorithm based on the social force model, the error of the abnormal situation is large because the change information of the scene is not taken into account. Therefore, VIF feature is introduced to represent time and space information of scenic spot perfectly.(4) A method of crowd abnormal recognition based on crowd tra-jectory information and convolution neural network is proposed. Firstly,the RLOF algorithm is used to extract the information of the crowd, and then the crowd information is clustered to obtain the overall trajectory information of the crowd. Different anomalies correspond to different trajectory characteristics. Based on this, the six anomalies are predefined,including abnormal aggregation, abnormal spread, straight run, ring run,cross run, and entrance gathering. At the same time, through studying the trajectory of the motion pattern, it can be found that the degree of bending of the trajectory of the crowd can be used as the basic basis for the deter-mination of the motion pattern. The trajectories are classified based on the curvature of the movement trail, and the abnormal behavior of the crowd is identified by the trajectory template of the predefined crowd.After recognizing the abnormal behavior of the crowd, we use the con-volution neural network to identify the video scene semantics.