Evaluation of YOLOv8n Performance for Real-time Human Detection on Autonomous Mobile Robots

This study presents the implementation and evaluation of the You Only Look Once version 8 nano (YOLOv8n) algorithm for real-time human detection on an autonomous mobile robot. The proposed system is designed as an edge-computing-based surveillance solution for monitoring restricted or difficult-to-access areas. The hardware platform integrates a Raspberry Pi 4B for visual processing and an Arduino Mega 2560 for navigation control through serial communication. Human detection is performed using a night-vision camera, while obstacle avoidance is supported by three ultrasonic sensors. A custom dataset was collected under various human postures, object distances ranging from 1 to 10 meters, and different lighting conditions. The YOLOv8n model was trained using 300 epochs with an image resolution of 640 × 640 pixels. Experimental results demonstrate that the proposed system achieves reliable real-time performance under varying environmental conditions. Under lighting variation tests, the model achieved 100% precision, 93.5% recall, 96.6% F1-score, and 93.55% accuracy with an average processing speed of 24.30 frames per second. Distance-based testing produced 100% precision, 92.42% recall, 96.06% F1-score, and 92.42% accuracy at 23.2 frames per second. Furthermore, autonomous navigation experiments confirmed that the robot was capable of simultaneously detecting humans and avoiding obstacles with response times ranging from 2.4 to 3.2 seconds. These findings indicate that You Only Look Once version 8 nano (YOLOv8n) provides an effective balance between detection accuracy, processing speed, and computational efficiency, making it suitable for deployment on edge-computing-based autonomous mobile robots.