Crowd Density Estimation And Application In Video Scene

Population density provide important basis for safety management and decision-making in public places.Due to the events such as traffic congestion,crowds in public places of large shopping malls or other large-scale activities becoming more and more crowded,the security issues become more and more serious.Therefore,the density of the population is very important for the public safety and crisis warning.We can pre-judge the crowd activities by estimating the population density.Considering the information provided by the estimation,we can strengthen the control of crows in case of emergency and prevent the crowd block in order to avoid large-scale crowd injury.Based on the commonly used algorithm of population density estimation,the convolution neural network model is used to detect pedestrians in the crowd.On the basis of pedestrian detection,the single plane positioning model is used to determine the pedestal's ground plane coordinates.And then estimating density.After obtaining the density distribution,we extended the application of density: population density high-risk warning and crowd evasion event detection.The main work of this paper is as follows:(1)pedestrian detection.In the video scene,the human body in the crowd is difficult to detect because of the occlusion.So we improve the occlusion model based on the depth learning pedestrian detection which is applicable to the crowd occlusion situation.After the deformation layer of the convolution neural network based on the depth learning pedestrian detection,we randomly set some units zero,calculate the mean,median,maximum and other statistical characteristics and calculate the statistical properties of its classification results to improve the crowd under the scene of pedestrian detection accuracy.Experiments show that the method can identify the pedestrians in the crowd more accurately,deal with the crowd in the block and environmental changes.(2)Mapping the pedestrian location.In the two-dimensional digital image,the pedestrian's position is not accurate due to the distortion of the camera and the crowding of the crowd,which is one of the reasons for the current population density estimation error.So the pedestrian's position on the ground plane in the natural scene is calculated according to the mapping between the digital image and the natural space.And the population density is calculated on the ground plane.First,the vision of the camera is corrected and the impact of the elimination of distortion is eliminated.And then the camera calibration is performed.After that,digital images in the pedestrian are mapped.Then we calculate the pedestrian position on the ground plane,count up the number of pedestriansin plane of per unit area for population detection to provide density distribution characteristics The Experiments show that the method can accurately locate the pedestrian position and improve the accuracy of population density estimation.(3)crowd incident detection.The change of density distribution provides a lot of information for the safety of the population.But the application area of density information is very narrow.This paper first demonstrates that the estimation of density can judge and forecast the danger of the population and provide important information for the safety of the population.And then we proposed that we use the density distribution of the vector changes,SVM classifier to detect the crowd incident.Experiments show that changes in density can alert the danger of population and identify the crowd in the video.