Font Size: a A A

Research On Deep Crowd Counting Algorithms Via Pre-Classification Of Density

Posted on:2019-05-11Degree:MasterType:Thesis
Country:ChinaCandidate:S Z WangFull Text:PDF
GTID:2428330563485960Subject:Bionic Equipment and Control Engineering
Abstract/Summary:PDF Full Text Request
This paper mainly focuses on crowd-counting algorithms for videos and images.We will introduces three methods for crowd counting,crowd counting algorithm based on regression model,crowd density estimation algorithm based on deep learning and deep crowd counting algorithm via pre-classification crowd density.Firstly,this paper introduces a crowd counting algorithm based on regression model.The foreground segmentation of the algorithm is used to segment the crowd from the scene,and the feature extraction part extract the handily craft feature from the acquired crowd area,and the crowd features are send to the regression model part for training.The nonlinear mapping relationship between the crowd characteristics and the corresponding number of images is established.However,the quality of the foreground segmentation algorithm will affect the final counting accuracy,and the handily craft crowd features are not robust to different scenes.In order to solve the above two problems,this paper introduces a crowd-counting method based on deep learning.This method uses a convolutional neural network to establish a nonlinear mapping relationship between the input image and its corresponding crowd density map.The sums of pixel value of the population density map is the number of people in the image.A single convolutional neural network is generally difficult to cope with dramatic changes in crowd density.We believe that the weight of a single network model is not compact.Based on the above motivation,this paper proposes a new deep crowd counting network algorithm via pre-classification crowd density.The algorithm pre-classifies the input images into several sets according to the number of people in each image,and their corresponding crowd counting networks are used for density estimation,which makes the algorithm be capable of coping with crowd gathering conditions in different scenes.Finally,in order to apply the deep crowd-counting algorithm,this paper uses PyQt4 to design a crowd counting system.We hope that the system designed in this paper can provide some inspirations and references for the crowd counting system design.
Keywords/Search Tags:Crowd Counting, Density Pre-Classification, Deep Learning
PDF Full Text Request
Related items