Pedestrian pose estimation using near set theory
Loading...
Date
2021
Authors
Haider, Muhammad Shangol
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Depending on the application type there are numerous stages of an automated computer vision system. Extracting informative foreground from the clutter of background data and identifying patterns are two stages of a typical computer vision system. Work had been done to use near set theory to identify spatially and descriptively near objects within an image. In this thesis, we extended this idea to videos and introduced the notion of temporal proximity for pattern recognition in spatial descriptive temporal domain. Each video is composed of a set of voxels which are the most basic elements of video defined as union of pixel and time stamp data. For extracting useful features, we observe the properties of topological spaces and introduce the concept of optical vortex nerve. Energy, area, 0-simplices (vertices), 1-simplices (edges) and 2-simplices (triangles) are some of the key features observed in this thesis. Finally, I explored pedestrian pose estimation as an application for applying all the concepts developed in this document in order to classify walking and running poses.
Description
Keywords
Content-based image retrieval, description, feature vector, probe function,closeness, near set, temporal proximity, maximal nucleus cluster, optical vortex nerve, persistence, proximity, shape, topological shape space