Research On The Key Technology Of Gait Recognition In Complex Monitoring Scene

Gait is a new biometric tool.Medical research shows that gait is a characteristic of both personalization and regularity.The gait has remarkable advantages,such as non aggression,non perception,far distance,and difficult to hide.With the expansion of China's infrastructure,monitoring equipment,a large number of large buildings such as banks,airports,subway stations and other important occasions often requires a large field of view,multi angle,and with the current domestic and international security situation continues to deteriorate,the monitoring environment is becoming increasingly complex.In this kind of complex monitoring scene,it is of great academic significance and application value to study the gait based target recognition and a variety of biometric fusion.Therefore,after analyzing all kinds of complex monitoring scenarios,we first divide them into single fixed view,multi view,large field of view,multi view and low resolution.The main work is as follows:(1)Aiming at some shortcomings of current gait feature descriptors in a single fixed perspective,we can get accurate,concise and efficient feature description by improving gait feature descriptors.Firstly,Gabor feature extraction,and then analyzed by PCA MMI,the two dimension space to generate gait characteristics,combined with the hidden Markov model to identify the target,and the effectiveness of the proposed feature extraction is verified by experiments in the walking state under different.Finally,a hierarchical classification scheme is proposed for the unknown single angle of view.Through this scheme,the calculation of the training model is simplified,and the recognition rate is improved.(2)In a wide range of monitoring scenes,many angle cameras are often needed to coordinate multiple angles with each other for no dead angle monitoring.On the one hand,multi view gait increases the amount of data,which is beneficial to recognition;on the one hand,multiple perspectives also bring huge uncertainty challenges.In view of the unknown angle of view existing in multi view scenes,a method of extracting the relevance features from the perspective is first proposed,and the validity of the method is verified by experiments.Then a prediction method based on regression analysis is put forward,that is,establishing a robust regression model through training,forecasting the unknown test perspective,and then proceeding according to the predictive perspective.Finally,the feasibility and efficiency of the classification recognition based on the visual angle prediction method in this paper is verified by the experiment.(3)In large field of view and low resolution scene,it often has the conditions to collect a variety of biological characteristics.It is easy to collect face and gait data in the image sequence.Therefore,based on these two biological characteristics,this paper develops a multi-Biometric Fusion study.Through the study of the fusion method,a more stable decision layer is selected to fuse two biological features of face and gait.Based on Bayesian theory,a Bayesian fusion strategy is proposed,and the effectiveness of the method is verified by experiments.