Research On Attitude Face Recognition Technology For Criminal Search System

With the rapid economic development and the acceleration of urbanization,the mobility of urban population in our country has greatly increased,and the crime phenomenon tends to be sudden and indeterminate.This has brought great inconvenience to police officers.In order to reduce the work of public security personnel,improve the efficiency of criminals search,face recognition technology is gradually applied to the field of security.In this thesis,we focus on the convolution neural network,which is a feature extraction algorithm of face recognition,and design a criminal detection system based on attitude and face recognition.Face detection and feature point location are the primary section of face recognition technology.In this thesis,we use a well-effective face detection algorithm based on NPD feature to determine the position of face region.Then,using AAM feature point localization algorithm locate the 68 feature points of face.Finally,the standard face image is obtained by using geometric and gray normalization to feature point position coordinates.The standard face image feature extraction,this thesis adopts the convolution neural network algorithm which has good robustness to illumination and attitude.The algorithm has two characteristics of sparse connection and weight sharing.The calculation method is similar to the animal’s visual system,in the field of image recognition has been successful.This thesis first introduces the basic theory and conception of convolution neural network.As the convolution neural network has a deep structure and a long computation time,it is not suitable for the real-time recognition task.Therefore,this thesis improves the original network structure by reducing the number of network layers and a variety of down-sampling methods.On this basis,Maxout function is used to replace the traditional activation function to achieve effective feature extraction.In this thesis,the network structure is designed by Caffe depth learning platform and tested under the LFW library View2 protocol.The recognition rate is 96.71%,and the feature extraction time is shortened.The feature extraction of a face image is only 104ms.Finally,the system is tested by collecting the face image through the camera.The recognition rate is 95.48%.To meet the criminals hunting system needs.At the end of this thesis,a criminal investigation system of face recognition is developed based on convolution neural network algorithm.The system collects images through the camera and performs face recognition on the acquired images to determine whether there are criminals in the surveillance area.And in this thesis,it includes the overall architecture of system,the network camera control,the face recognition,the data management and the alarm display.In the Windows development platform,using Matlab,C#and SQL Server database accomplish automatic search of criminals and finally it attains expected destination.