To address the issues of large blind spots and safety hazards during aircraft towing operations，a collision avoidance detection method integrating depth cameras and convolutional neural network（CNN） models was proposed. The CNN model was employed to automatically identify environmental targets，while the depth camera captured target distance information. The combined use of both enabled precise obstacle localization during the towing process. The trained CNN model and ZED2i binocular camera were deployed on the aircraft tractor，with communication facilitated via a CAN bus. Obstacle avoidance experiments were conducted in a test field. The results demonstrate that the constructed CNN model achieves a recognition accuracy of 0.911 and a recall rate of 0.803.Within a 10-meter ranging range，the distance measurement error remains below 0.3 meters，providing technical reference for collision avoidance safety detection during aircraft tractor towing operations.