Abstract Unmanned Aerial Vehicles (UAVs) play a vital role in applications such as aerial surveillance, disaster management , and urban air mobility. To ensure safe operation in dynamic environments, robust autonomous navigation and obstacle avoidance systems are essential. Conventional methods, which depend on predefined rules and sensor-based heuristics, often lack real-time adaptability in complex scenarios. To address this limitation, we propose a Deep Learning (DL)-enhanced Backtracking Search-optimized Customized YOLOv5 (BS-CYOLOv5) framework, which combines multi-sensor data fusion, real-time obstacle detection, and intelligent path planning. The dataset consists of UAV flight data gathered from RGB cameras, LiDAR, IMU, and GPS sensors. Pre-processing techniques, including normalization, are applied to improve model performance and generalization. For obstacle detection, the Customized YOLOv5 (CYOLOv5) model is utilized due to its high-speed inference and detection accuracy, enabling real-time obstacle recognition in diverse environments. Navigation and path planning are optimized using the Backtracking Search Algorithm (BSA), which dynamically adjusts flight paths to ensure collision-free and efficient trajectory planning. Experimental results demonstrate the effectiveness of the proposed approach, significantly improving obstacle avoidance accuracy and navigation efficiency. The CYOLOv5 model achieves an inference speed of 17.2 ms and a detection accuracy of 95%, while BSA ensures adaptive path optimization, minimizing collision risks. This research advances intelligent UAV systems by integrating cutting-edge DL and optimization techniques, enhancing autonomy, safety, and reliability in real-world operations.
Cherif et al. (Tue,) studied this question.