Research On Method And Application Of 3D Gait Recognition Based On Dynamic Time Warping Kernel

Gait recognition is a biometric technology that uses people's walking postures for identification.Compared with other biometrics,e.g.fingerprints,gait recognition has some unique and excellent features,such as long-distance recognition,no need to cooperate,difficulty in imitating or reproducing,strong confidentiality,etc.These features make gait recognition can be used to build a more convenient,safe and confidential biometric system,which has great application potential in the fields of security and intelligent monitoring.However,in real life,the gait of a person is easily affected by many factors such as walking speed,carrying,clothing,etc.,which makes it difficult to obtain a satisfactory effect for gait recognition.To further improve the performance of gait recognition methods and enhance the usability of gait recognition technology in real life,the main work carried out in this paper is as follows:1.A gait database with 3D spatial information is created.The database contained 152 subjects,each of which included 6 normal speed sequences,2 faster sequences,2 slower sequences,and 2 gait sequences carrying 3kg backpacks.Each gait sequence records four types of data: depth images,spatial transformation matrix,body index image,and skeleton model data.This is one of the largest gait databases in the world that contain 3D gait information.2.A three-dimensional gait feature representation method is proposed.First,each frame depth image is mapped into 3D space to form the 3D point cloud of the human body surface.Then,the point cloud is divided into several blocks according to a certain rule,and then the 3D spatial features of each frame of data are calculated in units of blocks.Finally,principal component analysis and multiple discriminant analysis are used to reduce the dimension of the feature vector to obtain the final representation of the gait feature.3.Gait recognition is performed using classifiers with the dynamic time warping kernel.The dynamic time warping algorithm is combined with the nearest neighbor classifier and support vector machine respectively,so that the two classifiers can accept gait feature vector sequences of different lengths to improve the recognition performance at different speeds.4.A real-time gait recognition system was constructed using the proposed method.The system uses Kinect to monitor the gait in real time,and once a complete gait cycle is received,the gait sequence is immediately identified.When the similarity reaches a certain threshold,the recognition result is displayed on the user interface and recorded in the database.