Research And Application Of Gait Recognition Algorithms For People In Coal Mines

In recent years,biometric identification technology has gradually developed and integrated into our lives,such as fingerprint recognition,face recognition,etc.have achieved good application effects,but gait recognition technology as the only method to achieve remote identity recognition,there is still much room for improvement.In some special applications,such as underground coal mines,the environment is characterized by dim light and a large amount of floating powder and coal dust in the roadway,which leads to the failure of face recognition and other identification methods that rely on high-quality image information.This thesis proposes to apply gait recognition technology to the specific scene of underground coal mine.In order to improve the accuracy of gait recognition and its applicability in coal mines,this thesis has mainly done the following work:(1)In view of the underground conditions leading to the traditional method of moving target detection effect is not ideal,and deep learning algorithm is overly dependent on a large of training samples,this paper puts forward a new kind of deep learning framework that uses background information to achieve static underground coal scenes when the number of training samples is small.First,the fixed background picture and the foreground picture are fed into the deep learning network as input information,and the simplified Goog Le Net Inception V1 structure is applied into the backbone network,and then the fixed background image information in the extracted surveillance video is used to guide the learning process of the network.In this way,the network can not only learn the characteristics of the object,but also learn the difference information between the object area and the non-object area,and finally the experimental results are analyzed through Intersection-over-Union(Io U),Recall,Precision and Average Precision(AP).The experimental results show that the network proposed in this paper can accurately extract the moving targets in the coal mine under the limited conditions of the small sample training set,which lays a good foundation for the subsequent extraction of gait features of the moving human body.(2)To solve the problem of incomplete extraction of usefu information and low recognition rate caused by single feature,a weighted fusion algorithm based on multiple features is proposed in this thesis.Firstly,three different dimensions of gait energy image,human contour,and skeleton are selected to express human motion information,and the three methods are optimized in a targeted manner.Specifically,the feature extraction of gait energy image based on PCA(Principal Component Analysis)is used to transform the gait energy map within a period into a one-dimensional vector,and the dimensionality reduction of the data is cariied out through PCA to retain the principal component data containing 95% information;based on the gait feature extraction of the wavelet descriptor of human contour,afer the wavelet description of boundary center distance is obtained,only 64low-frequency components are extracted to reduce the calculation amount of data and achieve the putpose of denoising;based on the skeleton feature extraction of contour and skeleton positioning,the improved method based on contour and skeleton was used to locate the coordinate of hip joint and obtain the lower limb model.The human body's moving gait features were represented by the changes of the four angle degrees between the line and the vertical line in the model.Then,the weighted feature fusion processing was carried out according to their respective recognition rates and experiments were carried out.Finally,the recognition rate is compared with the current popular gait recognition algorithm.The experimental results show that the multi-gait feature fusion algorithm proposed in this paper can achieve better recognition effect both in the open CASIA-B data set and the self-built MINE-GAIT data set.(3)Aiming at the specific environment of underground coal mine,this paper firstly proposes an improvement to the target detection method.Secondly,in order to improve the accuracy of gait recognition,a method of weighted fusion of the three gait features is proposed.Finally,a gait recognition system suitable for underground coal mine is developed based on the algorithm proposed in this paper,which can basically realize the intelligent indentification of personnel in the tunnel.The thesis includes 68 figures,13 tables and 80 references.