Persistent Person Identification and Tracking in Public Local Scenes using Unmanned Vehicle

  • Xiaoxiong Zhang

Student thesis: Doctoral Thesis

Abstract

Industries have reaped a lot of benefits from the applications of robots in the past few decades. Using robots reduces a lot of human effort and relieves humans from dangerous and cumbersome work. Their utilization has also led to a massive increase in the company’s productivity and profits. However, using mobile robots to assist humans in work and daily life is still in its initial stage. Fully autonomous person-following robots have great potential in many applications, including security surveillance, elderly care, and human collaboration. Even though person-following robots have been studied in many works, human intervention is required in most cases. In addition, normally the robots are designed to follow the target person in an easier situation where the appearance of the target is different from the surroundings. However, the person-following task becomes significantly more challenging when there are highly similar-appearing people nearby. This situation is frequently seen in places where people are in uniforms, such as schools, industries, hospitals, sports events, and the Gulf regions. The presence of similar-appearing people leads to tracking failure, causing the robot to follow the non-target person. With the rise of deep learning, robotics mixed with artificial intelligence have more potential for doing complicated tasks. This thesis proposes using a data-driven approach to design a fully autonomous robot system for person following in uniform crowds using visual data. The system consists of three components that work together to provide the robot with a high degree of autonomy and robustness. It involves the person identification system which takes place in the initialization step for the target identification, a novel tracking system that continuously locates the target using a visual object tracker, and a control system that maneuvers the robot’s movement. To address the challenges faced in the uniform crowd, we present a visual object tracker embedding the RGB and depth image. In addition, to fill the gap of lacking datasets covering uniform crowd scenarios and boost the development of the tracker addressing the challenges in the uniform crowd, this thesis collected a high-quality tracking benchmark dataset with uniform crowd scenes (The archiving and status of the dataset can be found in this website1). An intensive number of experiments have been conducted on the dataset and in real-time scenarios to compare the developed robot system and the state-of-the-art system. The results affirm the superiority and dependability of our developed system.
Date of AwardDec 2022
Original languageAmerican English

Keywords

  • Dataset
  • Face Identification
  • Face Verification
  • Ground Robot
  • Vision-based detector
  • Vision-based tracker
  • Real-time system
  • Robot Person following
  • Uniform Crowd

Cite this

'