Research on Multi object Detection Algorithm for Low Light Traffic

In intelligent driving systems, existing detection algorithms perform well on high-resolution and clear images, but the average detection accuracy decreases under low light conditions at night. This paper proposes a low light traffic multi-target detection algorithm based on YOLOv9 to address the issue of reduced AP in traffic target detection under low light conditions. Introducing PE-NET module in backbone: calculating and storing the residual of the original image through Laplacian pyramid, obtaining four different sizes of detail information; Using the LEF low-frequency enhancement module to capture and enhance low-frequency information in low light images of different sizes, reducing the impact of noise; By using the DPM detail processing module, information of different sizes is processed to obtain domain information of the image and edge information of the target, thereby improving the performance of traffic target detection. To validate the performance of the proposed algorithm, 5000 low light autonomous driving data were extracted from the BDD100K autonomous driving dataset for experimental verification. The experimental results show that for traffic multi-target detection under low light conditions, the improved YOLOv9 network has achieved a good improvement in detection performance compared to the original YOLOv9 network, with a 6.79% increase in mAp. Compared to the original YOLOv9 model, it is more suitable for driving in low light conditions.