Scenario Optimization Generation and Testing Method for Airborne Object Detection Algorithm

Artificial intelligence-driven object detection algorithms are extensively employed in the realm of unmanned aerial vehicles. However, the intricate and unpredictable nature of real-world application scenarios frequently poses challenges, resulting in detection inaccuracies in certain practical settings. Therefore, the urgent need arises for an efficient testing mechanism to assess the airborne target detection algorithm and identify scenarios prone to detection errors. In this paper, a novel scene parameter search approach grounded in the Markov chain Monte Carlo algorithm is introduced. This innovative method leverages scene parameters to meticulously define and model the environment within the UE5 platform, enabling to capture a diverse array of scene images for rigorous testing of the object detection algorithm. The experimental findings are noteworthy. The proposed approach demonstrates remarkable efficiency in identifying the majority of scene parameter combinations that trigger failures in the object detection algorithm, all within a limited number of searches. This result underscores the potential of the proposed method to significantly enhance the reliability and accuracy of object detection algorithms in unmanned aerial vehicles, paving the way for safer and more effective autonomous operations.