Study On Crowd Counting Method Based On Residual Estimation

Since the world has entered the new century,the economy has developed rapidly and the population has seen explosive growth.At the same time,as a huge amount of resources are in the hands of big cities,many people congregate in big cities.Large amount of people not only provide vitality for the development of big cities,but also bring many problems.For example,public safety problems such as stampede accidents caused by extremely dense crowds in public places such as concerts and sports events are one of the problems caused by high population density in big cities.In order to prevent these public safety problems caused by extremely dense crowds,a large number of researchers began their research work in the field of crowd counting,and crowd counting is gradually becoming a hot research topic.After years of development,the existing research work has been able to achieve high accuracy of crowd counting estimation.However,the problem of misidentification caused by the complicated background of the crowded scene in public places,the mutual occlusion in the crowds,the different densities of the crowds,and the huge differences in the sizes and scales of the pedestrians in the images are still the key issues in this research field.In this paper,we introduce a new method based on residual estimation in order to decrease the misidentification and improve the performance of the existing crowd counting methods,and then apply our idea to the methods based on density map to improve the quality of generated density map,which provides higher practical application value in the real scenarios.The main work of this paper includes:1)This paper briefly introduces the research background and significance of crowd counting,summarizes and analyzes the research status,respectively introduces some crowd counting methods based on detection,regression and density map,then gives a brief description of the research content of this paper;2)The theoretical knowledge and techniques of some methods related to this paper are briefly introduced,including the convolutional neural networks(CNNs)and supervised learning;3)A crowd counting method based on counting residual estimation is designed to improve the existing crowd counting method and obtain better performance in accuracy.We demonstrate our method on several crowd counting datasets to verify the improvement in accuracy of crowd counting estimation that the method based on counting residual estimation brings;4)We further apply the residual estimation method to the density map based method of crowd counting.We present a new concept called Residual Density Map,and design a network based on Residual Density Map.We combine our Residual Density Map Estimation network with the existing crowd counting method based on density map to improve the quality of the density map generated by the existing method.Experiments are carried out on the crowd counting datasets to verify that our method can improve the accuracy of crowd counting estimation as well as the quality of the density map.