Research On Effective Face Determine And Retrieval Model In Pedestrian Detection

Pedestrian detection and image retrieval are the key research directions in image processing.The research content has high theoretical level and great commercial value.It plays an important role in intelligent video surveillance,face verification system and automobile driver-assistance.In the video surveillance application scenario,it can produce a large number of surveillance data,and in the massive surveillance data often only a small amount of effective information,and these effective data is difficult to be found.In order to solve this problem,this paper advance the effective face determine and retrieval model in pedestrian detection based on pedestrian detection and face recognition technology.Through this model,only the pedestrian data with valid face is retained in the data collection,which provides a data base for pedestrian image retrieval.Firstly,the paper analyzes the pedestrian detection,effective face determine and image retrieval technology.The pedestrian detection method can be divided into the method based on background modeling and feature learning,Effective face determine can be divided into elastic model and geometric model,Image retrieval is divided into text-based and content-based.Through the comparison of the related technology,this paper adopts the method of feature learning in pedestrian detection,and combines the two decision models in the effective face determine,and use the content-based method in the image retrieval.Then,the paper analyzes the shortcomings of the current pedestrian detection model,and proposed the effective face determine and retrieval model in pedestrian detection.This model is divided into three parts,pedestrian detection,effective face recognition and effective face retrieval.In the pedestrian detection,the HOG feature is used,and the feature extraction method is optimized to speed up the detection speed.In the effective face determine process,propose distance symmetry of outline the key points and area symmetry of eye the key points,composed of double symmetry coefficient,through this to determine face posture,so as to achieve to the purpose of the invalid pedestrian data filtering;In the image retrieval process,the geometric position feature and the double distance similarity coefficient are proposed,calculate the distance between the key points and the center point of nose to obtain eigenvector,normalized eigenvector to get finally describe the features of faces,respectively use Euclidean distance and the mean cosine distance to calculate the distance between the vector,combined with the result of calculation with double distance similarity formula,The similarity coefficient between the images is calculated,and the image retrieval is realized according to the similarity coefficient.In traditional pedestrian detection,is often just detect pedestrians in the position on the image,but not to do further processing to detect target,judgment,so the effectiveness of the detection greatly reduced.This paper puts forward the effective face determine in pedestrian detection and retrieval model,by pedestrian detection,effective face determine,face feature extraction steps can be collected a large amount of data,such as pedestrians time information and the picture content.In order to fast and efficient use of these data,the model of image retrieval phase will be collected effective pedestrian database for similarity retrieval,get the pedestrian images with high similarity.Finally,the pedestrian detection was performed by collecting the test data,experiments show that the optimized HOG pedestrian detection method improved the pedestrian detection speed.The results show that effectiveness coefficient is set to 0.9 in the model,a better determine performance is achieved.In the pedestrian image retrieval,according to the double distance similarity to retrieve the first five similar images.The experimental results verify the rationality of the effective face determine and retrieval model in pedestrian detection,and can filter the invalid pedestrian data,improve the pedestrian retrieval performance,which has certain practicability.