Investigation And Application Of Face And Gait Recognition Technology

At present,biometrics recognition technology has attracted the attention of domestic and foreign security experts.Among them,face recognition has the advantages of non-contact and high security,but it is easily affected by factors such as facial expression,light intensity and occlusion.Gait recognition has the advantages of long recognition distance and difficult to disguise,but it is easily affected by factors such as Human emotions,weight-bearing,and clothing.Besides,a method combining gait dual-view fusion and regional down-sampling is proposed to improve the system's operating efficiency and robustness.The research content of this subject mainly includes:1.Pretreatment of face and gait.On the one hand,the Gaussian mixture model was used to extract the gait foreground of the video stream,the position information of the person was obtained through the alignment mode,and the final gait frame image is obtained through the normalization process.On the other hand,the Gaussian mixture model and the human body proportion method were used to extract the face image,and the final face frame image was obtained through normalization.2.Integrate research from the perspective of gait.Gaitset network model was used for gait recognition.The accuracy of gait recognition is 92.2% and 93.7% in two single-view angles of 36 degrees and 90 degrees,respectively.Addition fusion was performed using two-view gait,and the accuracy rate obtained by using the ranking method is 94.3%.3.Carry out bimodal fusion research on face and gait.Firstly,a bimodal biometric fusion algorithm was proposed to study the accuracy of facial and gait bimodal fusion and analyze the fusion results.The best results are obtained when the sequence length of the cosine distance between the face and gait is 11,and the correct rate obtained from the three different types of face and gait fusion of normal,backpack and clothing are 100%,100%,and 99.3%,respectively.Finally,the distribution of the three types of faces and the cosine distance of gait was explored,and the distribution analysis was carried out.4.Construction and optimization of the system platform.In the system design,a method of distinguishing people with a view in a 90-degree angle is proposed,which effectively and accurately records the time when pedestrians appear in the video.A regional down-sampling method used for quickly extracting faces and gait images is designed to improve the research and analysis efficiency,Furthermore,the basic framework of software and hardware,communication methods,and platform environment were introduced.Finally,the accuracy and performance of the system were tested.Theaccuracy of the live test reaches to 100%.The single face and gait recognition takes 0.165 seconds and 0.315 seconds respectively.