Detection Of Intensive Crowd In City Public Places

In the pedestrian detection research topic,automatic intensive population detection and estimation is very important.Including the detection of population density level,population statistics,crowd behavior testing.In recent years,although there have been many excellent algorithms in this field,it is still a challenging problem for automatic visual monitoring due to the diversity of monitoring sites,the complexity of monitoring scenes,changes in population scale,and so on.In this paper,the current population density estimation method is analyzed and studied.From the point of view of practical application,the population density estimation algorithm is divided into two categories:large viewing angle monitoring place and small viewing angle monitoring place,and according to the characteristics of each scene crowd,the efficient algorithm framework is put forward,And each gives an example validation.In the large viewing angle monitoring place,the camera is set up relatively high,the crowd is far away from the surveillance camera,the crowd's foreground,edge,characteristic points and so on are easy to extract,but they are often disturbed by some background features,which have great influence on the accuracy of population density detection.In this paper,we propose a feature extraction method that uses background mask to mask background interference and fuse a variety of local features,and use SVR to establish population regression model to analyze the population.Compared with the previous methods,the foreground mask algorithm can effectively shield the interference of background features,and the fusion of multiple features can keep the stability of the algorithm when the crowd is disturbed by background interference.In the Grand Central Station dataset,using the algorithm,the MAE and MRE metrics were reduced by 6.54 and 21.9%,respectively,improving the accuracy of the test.In the small viewing angle monitoring place,the camera set up a low angle and close to the crowd,the crowd details are too serious.When the population density is high,the crowd is crowded,leading to serious shelter.For this scene,this paper improves the traditional population density detection algorithm based on texture,The monitoring scene is first corrected and divided into multiple sub-regions.Only when the number of sub-area foreground pixels exceeds the set threshold,the texture feature analysis is continued for the area,and the population density of each sub-area is counted.In the subway crowd video,using this method,the average per-frame processing time is reduced by 0.104s.