FOLLOW-ME AN AUTONOMOUS VISION-BASED TRACKING SYSTEM by

Abstract

Drone usage has skyrocketed over the past years, as these vehicles are being used for more complex and innovative tasks. This project covers the creation of “Follow-Me”, an autonomous quadcopter that is capable of tracking and following users using computer vision. This drone was constructed from the bottom-up, requiring a quadcopter frame, motors, ESCs, an IMU, and batteries. Both a microcontroller and a microprocessor were incorporated, each serving a unique and important role in the embedded system. Establishing stable flight was the first step after assembling the quadcopter, explored using PID controllers to correct rotational error along each of the three directional axes. To detect the user being tracked, computer vision methods were researched and implemented, achieving full-body detection, face detection, and head angle calculation. A cloudcomputing model was utilized to enable these computationally intensive algorithms to be run externally, satisfying low-latency requirements. The results from detection were integrated with the system to achieve autonomous tracking. Specifically, the positional error of the user was computed in each frame, enabling the drone to re-position itself appropriately. Despite some shortcomings, Follow- Me accomplished its goal of basic tracking and following.