The Development of a UAV Target Tracking System Based on YOLOv3-Tiny Object Detection Algorithm.

Unmanned Aerial Vehicle (UAV) machine vision target tracking has a wide range of applications. The most important thing in machine vision target tracking is the real-time performance of the object detection algorithm. At present, the amount of computation of the mainstream target detection algorithms used is too large. The processing speed of the target detection algorithm to the image collected by the computer is too slow, which makes the UAV lack real-time tracking. Therefore, it is very necessary to use a fast object detection algorithm in machine vision target tracking in UAVs. YOLOv3-Tiny is an object detection method based on deep learning. YOLOv3-Tiny optimizes the network structure based on YOLOv3 and reduces the output of one scale. In this work, we use human images to train YOLOv3-Tiny to make it can recognize humans as tracking targets. Then, we use YOLOv3-Tiny to analyze the images collected by the UAV for target tracking. Finally, the control signal of the tracking system is output to the flight controller of the UAV. For the tracking effect of the system designed in the research, we tested the tracking system in a complex indoor environment. It is found that the UAV indoor target tracking system based on YOLOv3-Tiny can effectively achieve target tracking.