Research Of Dense Crowd Image Analysis And Crowd Counting Based On Convolutional Neural Network

With the continuous development of the world economy and the continuous improvement of medical level,the population scale is expanding constantly.Meanwhile,the issues of dense crowd security are also becoming more and more serious.Malignant events caused by population aggregation occur frequently all over the world.Therefore,how to monitor and analyze the dense crowd effectively has gradually become a heated research topic.As an important branch of artificial intelligence field,deep learning has achieved breakthroughs in various fields so that computer vision,which is an important new field of deep learning,has also achieved great achievements.This paper analyzes the research status of dense crowd counting based on these backgrounds,introduces and compares the advantages and disadvantages of various methods,and proposes a scheme of dense crowd counting based on convolutional neural network.The specific content of this paper consists of the following three aspects:(1)This paper proposes an input adaptive multi-channel convolutional neural network model structure based on deep learning.This network model can be well adapted to the input images with different sizes.Moreover,the multi-channel network model with different receptive fields is used to analyze and process the input image,which improves the overall performance of the model and the accuracy of population estimation in the meantime.(2)A new large-scale dense crowd image dataset TJU Crowd Dataset is constructed,which consists of 400 high-resolution static images.Many scenes in the dataset almost cover all the possible scenarios in everyday life,and each image contains a large number of people with very high density.For each image,the author has made an accurate mark,which has a total of 208,180 personnel mark information that appears in the dataset.(3)The scheme of dense crowd counting presented in this paper can be verified by the experiments,which are carried out on the TJU Crowd Dataset,UCF CC 50 data sets,and UCSD data sets.Besides,the experimental results are compared and analyzed with the previous research.The experimental results show that the densecrowd counting scheme based on convolutional neural network proposed in this paper can achieve better accuracy in dense crowd counting.