Autonomous Control with Vision and Deep Learning: A Raspberry Pi Edge Computing Platform for Obstacle Detection in SUAV Path

The article addresses the critical need of obstacle detection for avoidance in the navigation path of the Small Unmanned Aerial Vehicles (SUAVs), particularly in complex environments with varying obstacles. Path planning and navigation for SUAVs is complex due to the variability of the navigating environment and the presence of diverse obstacles at low altitudes, particularly in populated areas. In this article, we propose obstacle detection in SUAV flight paths using state-of-the-art convolutional neural network (CNN) based object detection algorithms, particularly the YOLO (You Only Look Once) family, deployed on Edge devices such as Raspberry Pi. The study involves a comprehensive experimental evaluation of YOLO models for accuracy and speed in real-time obstacle detection. The methodology includes dataset acquisition, model training, and deployment on Raspberry-PI being a single board computer acting as a host using OAK-D camera. Results demonstrate the efficacy of these AI-on-the-Edge solutions in enhancing UAV navigation safety through autonomous obstacle detection for real-time avoidance. The findings highlight the feasibility and performance of integrating CNN-based vision systems with Edge computing for UAV applications in dynamic environments. Comparing the YOLO family, we achieved the highest Precision, Recall, and Accuracy for the YOLOv6.