Real-time Multi Pose Trajectory Tracking based on OpenPose Keypoints
Adam Surówka · 2021 11th IEEE International Conference on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications (IDAACS) · 2021
Currently, there exist several open-source computer vision libraries designed for human pose estimation from photos and videos. They are mainly focused on the possibility of detecting individuals in the image and returning their skeleton determinants. An often overlooked or underdeveloped functionality is tracking the trajectory of the detected people, which presents a serious problem in the process of design automated video surveillance systems. In this work, the author attempts to develop an algorithm for tracking multiple human poses in real-time, based on simple decision filters. The developed solution is designed to work with keypoints obtained from a selected open-source human poses recognition library. The author reveals details related to the method of processing and analysing obtained keypoints, describes the concept of decision filters and presents the results of the software implementation.