University of Twente Student Theses

Login

Towards view-invariant Gait Recognition from monocular video based on Human Pose Estimation

Bousias, Dimitrios (2021) Towards view-invariant Gait Recognition from monocular video based on Human Pose Estimation.

[img] PDF
13MB
Abstract:Gait recognition is the biometric method that can differentiate and identify individuals by the way they walk. Gait as a biometric feature has some interesting characteristics as it can be collected at distance while it can be very hard to fake. Previous gait recognition works that rely on human silhouette representation, are often dependent on robust contouring or background extraction methods. Additionally they can be very limited regarding the viewing angle or require specific conditions to be met. Inspired by recent progress in the field of human pose estimation and skeleton based gait recognition methods, we propose a framework for extracting markerless motion capture data from monocular video and identifying individuals based on extracted features. The generalizing power of off-the-shelf pose estimators towards in the wild videos is tested. The approach is aiming towards a view-angle and clothing invariant solution. A gait dataset is acquired to validate the uniqueness and permanence of various gait features. We report results of verification and identification experiments which are compared to respective ones attained with a commercial depth sensor. Correct identification rates of 79% up to 88% in the overall experiment are achieved using different combinations of features and template matching methods. Possible shortcomings of the method related to view-angle dependent bias or the filtering of identity information are discussed.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Electrical Engineering MSc (60353)
Link to this item:http://purl.utwente.nl/essays/86202
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page