Font Size: a A A

View-invariant Gait Recognition Based On Kinect Skeleton Information

Posted on:2018-01-18Degree:MasterType:Thesis
Country:ChinaCandidate:Y F WangFull Text:PDF
GTID:2348330512984431Subject:Signal and information systems
Abstract/Summary:PDF Full Text Request
Gait recognition is a kind novel biometric identification technology,which takes advantage of people's walking style to identify the person.The most different place comparing with other biometric identification technologies is that it does not require subject cooperation due to its non-invasive capturing process.Moreover,as gait is a kind of behavioral biometric feature,it's natural and hard to imitate which just like people's handwriting.Based on these advantages,gait recognition has an extensive application future in access control,human-computer interaction,medical diagnosis and so on.Gait recognition will show more powerful along with the development of algorithm and speed of calculation.Gait recognition techniques at the state of the art can be roughly divided into 2D-video based and 3D based approaches.The 2D-video based methods are generally convenient to apply,however most of them are easy to be affected by the covariant such as illumination and view.The 3D based methods are innately able to against the view changing,but it's complicated to achieve calibration in some multi-camera methods.Methods which utilize the depth camera can avoid the camera calibration,however the depth camera seems too expensive to expand in the practical application.In recent years,since the Kinect released in 2010,the depth sensors are tending to be more commercial and economic.Kinect is barely affected by illumination,and it's convenient to make use of depth information to divide the people from the background.Moreover,it's easy for Kinect to track and provide the 3D coordinate information of skeleton joints.Therefore Kinect has been used as a new tool for gait recognition.The main innovation and contribution of this paper are showed in the following two aspects:(1)We take advantage of Kinect(second generation)to establish a new kind 3D skeleton based gait database which is composed by 52 persons in 6 predefined and 4 user-defined view directions,totally 1040 walking sequences.The 3D position information of 21 skeleton joints at each frame and corresponding 2D silhouette images are included in each walking sequence.(2)We propose two kinds of view-invariant gait feature according to the 3D information of joints.We select 8 skeleton length as the static feature and 8 dynamic angles of swing limbs as the dynamic feature,which are verified to be view-invariant.Specific to the dynamic feature,we also discuss some key problems such as period extraction and similarity measurement.Based on this,we make a feature fusion on the matching level and get the classification result with NN(Nearest Neighbor)classifier.Finally,based on proposed database,the recognition result of the static,dynamic,and feature fusion are inspected respectively in the case of stable view and changing view.In addition,we compare our methods with traditional 2D-video based and other Kinect based methods.We also testify our method on another Kinect based dataset.The experiment results show that the proposed method has a good recognition performance.It is also noticed that our database has a potential for providing a bridge between 2D and 3D methods.
Keywords/Search Tags:biometric identification, gait recognition, Kinect, gait database, feature fusion
PDF Full Text Request
Related items