A YOLOv9-Based Night-Time Signal Light Detection Method in Carla Environment

With the continuous progress of machine learning technology, automatic driving perception technology is gradually becoming mature. As a guarantee of normal traffic operation, signal lights can effectively guide vehicles to pass correctly. The recognition of signal lights mainly relies on pure visual target detection technology. And the correct recognition of traffic signals can reduce the occurrence of traffic accidents and provide safety for unmanned driving. However, the recognition effect of small targets at night is poor, and is prone to misdetection and omission, etc. Therefore, the pure visual target detection technology for traffic signals still faces great challenges to accurately identify targets in special scenes. In this paper, the latest YOLOv9 image target recognition technology is used to improve the target detection ability in harsh environments. The advantages and disadvantages of different algorithms in low illumination environments such as night time are evaluated by using YOLOv8 and YOLOv9 target detection algorithms. Carla simulation environment was conducted to validate the recognition effectiveness of YOLOv9. The results show that the YOLOv9 algorithm is better at recognizing signal lights at night and the detection accuracy exceeds that of the previous versions of the series, ensuring the recognition accuracy of signal lights in different environments.