Visual-based Construction And Analysis Of Crowd Heat Maps In Scenic Spots

With the rapid development of tourism and the need of building "smart scenic",various problems in tourism management are faced.For example,we should know how to get the heat of the crowd in order to timely take appropriate measures to ease the flow of passengers,we should know what kind of attractions the tourists like in order to develop the marketable attractions,we should analyze the crowd behavior in the spots in order to prevent the occurrence of vicious incidents such as stampede,violence,etc.These are the difficulties that need to be overcome urgently in the transformation of the traditional scenic spots in China.Therefore,the research on the construction of crowd heat map and the analysis of crowd behavior has become the key research direction.Summarizing the deficiencies of domestic and foreign studies,on the one hand,the density map estimation research usually chooses the unified convolution neural network model to carry on the single picture density estimation,and the uneven density distribution image sequence considers insufficiently;on the other hand,the crowd behavior analysis studies uses the movement tracking method,however in the complex environment,tracking is more difficult.In view of the above two aspects,this paper proposes a density pre-classification of the thermal map to build the convolution neural network model,and based on the heat map of crowd flow,crowd behavior analysis.The main work of this paper is as follows:1)A heat map based on density pre-classification is proposed to construct a convolutional neural network model.Many researchers often choose a uniform convolutional neural network to estimate the depth of crowd density estimates.However,when dealing with such video frame images of scenic spots,it is difficult to train a compact network when the crowd density is unevenly distributed.To solve this problem,this paper proposes a density pre-classification method to improve the compactness of the heat map construction network.By improving the network structure,a crowd heat map estimation algorithm based on convolutional neural network is proposed,and the accuracy of our method is verified through experiments.2)Research on the flow of people in scenic spots based on the spatial and temporal interaction method of heat map.Analyze the population distribution at different times in the scenic spot so that the scenic spot managers can make the right choices in time.3)A crowd behavior analysis method of macro-group relationship model was proposed.According to the characteristics of crowd behavior in the heat map,the mathematics modeling method is used to calculate the crowd clustering area in two frames based on the previously constructed heat map.Through the change of similarity,a macro-group relationship model is constructed to study the relationship between groups in the image,so as to analyze the crowd's behavior such as expansion,disappearance,and detention.