Indoor Pedestrian-Following System by a Drone with Edge Computing and Neural Networks: Part 2 - Development of Tracking System and Monocular Depth Estimation

This paper is the second installment in a series on indoor drone pedestrian tracking utilizing edge computing and neural networks. Building upon the SLAM and EKF technologies introduced in Part 1, this paper introduces Monocular Depth Estimation to reduce camera costs and overall weight. The system leverages AI-driven depth information for indoor positioning and real-time human tracking. Experiments demonstrate the drone's ability to autonomously track a specific individual indoors using vision and IMU sensors. Key contributions encompass an AI-based tracking system employing YOLO v3 and a novel depth estimation approach that supersedes traditional depth cameras.