Research On The Algorithm Of Crowd Desnsity Detection

As rapid development of social economy increasingly enriches people's daily life,the circumstances of people-gathering become more and more prevalent.As a result,stampedes occur frequently.Since the technology of crowd density detection can solve the problem effectively,it has attracted more and more attention in academia.This paper presents two brand-new crowd density detection algorithms aiming at the shortcoming of existing algorithms.One is based on the support vector machine?SVM?.It combines distinct texture features,which are extracted from images at different levels into a more powerful feature descriptor based on image pyramid.And then a model is trained with the support vector machine.Taking advantages of deep learning,an improved convolutional neural network crowd density detection algorithm is proposed in the paper.It merges a diversity of information with distinct concepts which is based on the traditional texture features in the joint network to improve the robustness of algorithm.On the open dataset PETS₂009,the results show that the algorithm based on support vector machine of multiple features is less affected by illumination,and has higher accuracy than the general algorithm.The results of improved convolutional neural network?CNN?show significant performance even in low image resolution.It is more suitable for practical application demand.In general,a large number of experiments demonstrate strong practicalities as well as a broad application prospect of the two crowd density detection algorithms.