Vision-Based Identity Aware Person Following Robot in Uniform Crowd

Abstract

Person-tracking robots have many applications, including security, surveillance, elderly care, autonomous driving, and socializing robots. The tracker plays an important role in those robots where the appearance of the target is the highly discriminating feature used for the separation of targets from the background. Such a task is easier when the target has a remarkably different appearance than the other people in the environment. However, the task becomes significantly more challenging when there are highly similar-appearing people. The uniform crowd is frequently seen in places like schools, industries, healthcare, and sports events, which are commonly observed in the Gulf regions and Asian countries. The presence of a similar-appearing distractor leads to tracking failure, causing the robot to follow the non-target person. This work primarily aims to develop a robust robot system that can follow a specific target person in a uniform crowd. The system has several components that work together to provide a high degree of autonomy and robustness. It involves the person identification system, which ensures the identity of the person; a novel tracking system that tracks the target person robustly in the uniform crowd; and a control system that controls the robot's movement. Despite the abundance of uniform crowds in many contexts and the challenges they exhibit, there was a lack of video datasets dedicated to benchmarking tracking algorithms in such contexts. A high-quality tracking benchmark dataset with uniform crowd scenes has also been developed (The archiving and status of the dataset can be found in this website1). Also, an intensive number of experiments have been performed with datasets and in real-time to affirm the superiority and dependability of the developed system. The developed system has tracking speed double (70+FPS) compared to other SOTA system (30 + FPS) and also has robust performance compared to all the developed system.

Date of Award	Jul 2022
Original language	American English