<p>Unmanned aerial vehicles (UAVs) equipped with advanced vision and intelligence techniques have a wide range of real-world applications. This paper presents a framework for an autonomous detection and avoidance (DAA) system for UAVs that utilizes deep learning and transfer learning. The system addresses the challenges of aerial collision avoidance by integrating object detection, tracking, and trajectory planning. We employ the YOLO model for real-time object detection, paired with the DeepSORT tracker for monitoring object trajectories. To enhance the performance with limited training data, transfer learning techniques are utilized, allowing the model to adapt previous knowledge to new tasks. The Frenet coordinate system is employed for trajectory planning during obstacle avoidance. Simulation tests in realistic environments demonstrate the system’s effectiveness; it achieves a high accuracy and minimizes the need for extensive physical flight tests. This work provides a robust approach to ensuring safer and more efficient UAV operations in complex scenarios.</p>