Research On Crowd Counting Method Based On Deep Learning

Crowd counting is a key and fundamental issue in the field of public safety.With the explosive growth of image data and the rapid development of deep learning technology in the field of computer vision,the crowd counting or crowd density estimation based on deep learning technology is a key research field and has a high practical application value.This paper mainly uses the latest theory and technology in deep learning to study crowd density estimation,the study has theoretical and practical value,the main works are as follow:Firstly,the traditional methods of crowd counting and the method of crowd counting based on deep learning are studied,and the advantages and disadvantages of the traditional methods and some of the crowd counting methods based on deep learning are summarized and compared,and two adaptive Gaussian kernel density estimation method are put forward.Secondly,a multi-column multi-task convolution neural network(MMCNN)is proposed for the shortcomings of some of the methods based on deep learning and combined with the advantages of each method.The network model is a "map-to-map" and "end-to-end" system,the input data can be arbitrary resolution picture,the network's predictions are in the corresponding size of input pictures.Experiments show that the performance of MMCNN model is superior to other methods based on deep learning.Then,based on the shortcomings of MMCNN and the latest deep learning theory and techniques,an improved multi-column convolution neural network(IMCNN)model is proposed creatively.Based on the advantages of MMCNN,this model not only overcomes the difficult problem of training,but also improves the fitting ability of the network.Experiments show that the IMCNN model is effective and efficient,and has practical application prospect.Finally,this paper puts forward some practical applications of network's prediction,such as: treating the density map as deep visual features,sub-regional crowd counting,crowd density grading display,analysis of the trend of crowd count and high crowd density detection.The experimental results show that the crowd counting method based on deep learning has a good practical application prospect.