Research And Application Of Pedestrian Detection Based On Head And Shoulders Feature

Pedestrian detection technology has always been a hot and difficult point in the field of machine vision.It provides an important technical guarantee for intelligent monitoring,vehicle driving and construction of intelligent city.In recent years,convolution neural network has made many breakthroughs in the field of image processing.It extracts image features directly from the image pixels by simulating the learning process of human brain,and combines feature extraction and classifier to a learning framework to classify and identify the related objects.In this thesis,a pedestrian detection model based on convolution neural network is designed on the basis of full study of pedestrian detection and convolution neural network model,which is applied to pedestrian detection.1.The head and shoulder of the pedestrian is not easy to block,with stable contour feature.Therefore,the pedestrian head and shoulder as detection object,to reduce the detection range.According to the fixed proportion of the head and shoulder parts,an improved algorithm is proposed for detecting the head and shoulder parts of the human body based on the frame difference method based on edge detection and the mixed Gauss model.The edge extraction and the inter frame difference method are combined to extract the edge feature of the continuous image,and the image with edge features is divided into three frames.Secondly,the hybrid Gauss extraction is carried out for the continuous highlight.Finally,the difference between the two methods is calculated,and the result is denoising through two valued and morphological processing to get the head and shoulder part of human body.2.when the Le Net-5 convolution neural network is used to identify the head and shoulder parts of the human body,there is a problem of poor convergence,which is caused by the size of the convolution kernel and the number of the network layers.The influence of the number of neural network layers and the size of convolution kernel on the recognition effect is analyzed systematically,and a convolution neural network structure is designed for the head and shoulder detection of the human body.Through many experiments,we can get that when the number of layers of convolution neural network is 7 layers and the size of convolution kernel is 5 × 5,the recognition effect of the model is the best.3.Though extracting and using the convolutional neural network to classify thehead and shoulder part of pedestrians,we design a pedestrian detection system.Through the experiment comparison,this method not only can reach 96% accuracy rate,at the same time reduce the detection time,also can avoid the error caused by the body occlusion,and have a better detection effect for the pedestrians riding bicycles and electric cars.