Font Size: a A A

Gait Analysis And Recognition Based On Kinect Bone Information

Posted on:2021-04-18Degree:MasterType:Thesis
Country:ChinaCandidate:G C GuoFull Text:PDF
GTID:2428330629480397Subject:Electronic and communication engineering
Abstract/Summary:PDF Full Text Request
The rapid development of modern society has pushed humans to pay more attention to security issues.Biometric identification,as an identification technology required for places with high security sensitivity requirements,has attracted widespread attention from computer researchers.In the field of intelligent video surveillance,traditional biological features such as human faces?irises and fingerprints can be well recognized under contact or close range,but cannot be recognized in long-distance scenes.Gait,as a perceptible biological feature,has the characteristics that are not possessed by traditional features such as easy acquisition,long-distance recognition and difficult camouflage,Therefore,using human walking posture to analyze human body in the field of video surveillance to identify human identity has become the current research hot topic.Gait recognition methods are mainly divided into two-dimensional information and three-dimensional methods.The gait recognition algorithm based on two-dimensional images is complex and easy to be affected by light,human body wear and angle of view changes.The gait recognition algorithm based on three-dimensional image has low complexity and is not easy to be interfered by external environment factors.It can obtain 3D skeleton information of human body and realize background segmentation of human body.In this thesis,Kinect depth sensor is selected as the three-dimensional information collection equipment,and the color image and bone information of human body are used as the input data.The improved new DTW algorithm and k-nearest-neighbor algorithm are used to identify the gait.A gait analysis and recognition method based on Kinect bone information is proposed.The main process of gait recognition consists of three parts: data collection and preprocessing,dynamic and static feature extraction,and recognition algorithm.The main research contents of this article are as follows:(1)Data collection.This thesis uses UPCV(University of Patras Computer Vision)datset to verify the effectiveness of the proposed method.Since the UPCV dataset lacks two-dimensional images,this thesis uses Kinect V2 to collect color and bone images of 20 students in the laboratory.A total of 20 students participated in the self-built dataset,Each person walked five gait sequences at normal speed at 0 degrees(parallel to the X axis)and 30 degrees(positive to the X axis)and there were 200 gait video sequences in total.Each gait sequence contains the corresponding color image of each frame and the three-dimensional skeleton coordinates of 20 skeleton points of each frame.(2)Gait feature extraction.Using human physiology structure,five human skeletal modules(spine?left arm?right arm?left leg?and right leg)are proposed to construct a human body model.After obtaining three-dimensional skeletal coordinates of the human body and preprocessing,this thesis extracts from many aspects that can represent the individual Gait features,including static features and dynamic features;Based on the 3D skeleton information collected by Kinect,7 static features such as thigh length and 3 dynamic features such as knee joint angle are extracted.For the dynamic and static features,the similarity measure and the average of the static length in a gait cycle are used to prove the feasibility of dynamic and static feature extraction.(3)Gait classification and recognition.The improved new dynamic time warping algorithm and the nearest neighbor and K nearest neighbor algorithm are used to achieve gait recognition comparison using three gait feature methods: static gait feature,dynamic gait feature,static gait feature and dynamic feature fusion.According to the proposed gait feature fusion recognition method,the distance calculated from the dynamic and static gait features is used as the matching score of each corresponding feature.Finally,the total matching score of the test sequence and training sequence is obtained by using the fusion algorithm of the improved weighted addition for the matching score of the dynamic and static skeleton features.so that the classification and statistical effects of gait recognition are completed using NN and KNN algorithm.The 90% and 92.5% recognition results achieved in UPCV database and self built data,the recognition rate is about 20% higher than that of two-dimensional image recognition in the same data set and comparison literature,which verifies the feasibility of this method.Finally,a gait recognition simulation demonstration system was implemented and developed by using related technologies such as OpencnCV and Kinect SDK and the method proposed in the thesis was integrated into the system.
Keywords/Search Tags:Kinect, dynamic time warping, feature fusion, gait recognition
PDF Full Text Request
Related items