Boosting Unmanned Aerial Vehicle (UAV) Navigation Capabilities: Improving Obstacle Detection and Evasion using the You Only Look Once (YOLO) Algorithm

This paper investigates an improved approach for real-time obstacle detection and avoidance in unmanned aerial vehicles (UAVs) using the latest version of the You Only Look Once (YOLOv8) object detection algorithm. While traditional obstacle avoidance methods often rely on sensors such as LiDAR, stereo cameras, or feature extraction techniques like Scale- Invariant Feature Transform (SIFT), these methods can be hardware-intensive and less adaptive in dynamic environments. The proposed method leverages deep learning through convolutional neural networks (CNNs) embedded in YOLOv8 to detect and localize obstacles with high speed and accuracy. To enhance evasive manoeuvres, the system integrates a dimension augmentation algorithm that assigns zone-based class weights to prioritize obstacle significance, influencing UAV navigation decisions in real time. The model was evaluated in a controlled environment to ensure precision in initial testing. However, challenges remain in extending performance to complex outdoor scenarios, which are discussed. Results demonstrate YOLOv8’s superiority in detection accuracy and processing speed compared to earlier models. The proposed approach provides a low-cost, sensor-light alternative that improves UAV autonomy. This work contributes by integrating a novel weight-based decision framework with YOLOv8 for UAVs, offering a scalable method for future real-world deployment.