Design and Implementation of a YOLOv5-Based Drone Campus Object Recognition System

Abstract

Addressing the issues of low efficiency and limited coverage in traditional campus security patrols, as well as the fact that drone patrols can only collect data and lack real-time target recognition capabilities, this paper designs and implements a drone-based campus target recognition system based on the original YOLOv5 model.Developed using a cross-technology stack architecture comprising Vue, IDEA, and PyCharm, the system is divided into three core modules: image recognition, video recognition, and real-time video stream recognition. Without requiring algorithmic modifications or retraining of the YOLOv5 model, it achieves efficient recognition of campus scene data collected by drones through engineering integration techniques.Test results demonstrate that the system can reliably identify typical campus targets such as pedestrians, vehicles, and buildings. The image recognition accuracy reaches 85.2%, the video stream processing frame rate is maintained at 15–20 FPS, and real-time recognition latency is controlled within 450 ms, meeting the practical application requirements for scenarios such as campus drone patrols and security monitoring.This system facilitates the rapid engineering implementation of mature object detection models, providing a lightweight, highly available solution for smart campus security infrastructure.