Academy & Industry Research Collaboration Center (AIRCC)

Volume 12, Number 07, April 2022

A Pose-based Image Searching using Computer Vision and Post-Estimate


Hang Wang1 and Yu Sun2, 1University of California—Berkeley, USA, 2Cal Poly Pomona, USA


As the cost of human forces increases, people in some careers, like the artists, may find some difficulties when models are needed in the processes of making art. Obviously, one alternative solution is to find pictures online, however, when some specific poses are needed, they may also find some difficulties to describe them. This paper develops an application to search for pictures with the target pose described by the user's graphical input. In this project, one hundred images describing distinct actions and activities with various poses and view-angles are collected. Mediapipe is then used to analyze those images in a quantitative way. We also embedded a User Interface that allows the user to imitate the intended pose as well as the viewing angles by simply dragging around body joints of the figure. Different sets of feature points and matching algorithms are also tested to find out the best solution.


Computer vision, Pose detection, Image searching.