Loading...

Human Identity Recognition Through Gait and Body Motions Analysis

Jebraeeli, Vahid | 2021

177 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 54955 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Ghaemmaghami, Shahrokh
  7. Abstract:
  8. Among all biometric approaches, gait analysis is one of the most practical methods for human identity recognition. Gait has a lot of advantages over other biometrics like face recognition, iris recognition, fingerprint, etc. First and foremost, the gait data can be collected from a distance, and there is no need for subject’s cooperation. Another advantage of this biometric method is its cost-effectiveness and the fact that it does not need high-resolution images. But there are significant challenges in detecting and analyzing this feature. One of the most important challenges is decreased recognition accuracy caused by identity-irrelevant factors like camera viewpoint and changes in walking conditions. Researchers have proposed different solutions for this problem and the recent ones are mostly ML-based approaches. GaitSet, GaitGAN, and GaitPart are three examples of the most recent networks which have reached the rank-1 identity recognition accuracies, respectively, equal to 73.3%, 84.2%, and 88.8%. But still, the common problem of these networks is their low recognition accuracy in the presence of identity-irrelevant factors, especially the changes in camera angle.In this thesis, we have first made improvements in the GaitSet network to make it compatible with our general algorithm, and then, we have proposed a bigger network in order to disentangle the overall algorithm from camera viewpoint factor. Our proposed model consists of three parts. The first part is the view transformation network, a generative network that can transform the initial gait image into 10 different images from different viewpoints by preserving the identity information of the initial image. The second network is the improved version of GaitSet which can also be replaced by GaitGAN and these networks are responsible for recognizing the identity of 11 gait images (one initial and 10 transformed images), and at the end we make a decision on the identity of the initial input image by utilizing a weighted majority voting ensemble approach. Our proposed model could reach an average rank-1 identity recognition accuracy equal to 89.4% over all the data from the famous gait dataset named CASIA-B. This is the best gait-based identity recognition accuracy ever reached. Moreover, a statistical analysis of our model results shows that our proposed model's recognition accuracy is significantly resilient to identity-irrelevant factors like changes in the camera viewpoint and walking conditions.
  9. Keywords:
  10. Biometrics ; Deep Learning ; Identification ; Body Poses ; Gait Analysis ; Body Motions

 Digital Object List

 Bookmark

No TOC