The low-light condition brings huge challenges to the traffic object detection, limiting its application in complex scenarios such as the round-the-clock traffic monitoring and automatic driving. To address this issue, an accurate traffic object detection method in low-light conditions is proposed. First, the dark channel prior knowledge is utilized to restore the scene radiance via the transmission map and the estimation of the atmospheric light. Meanwhile, the grayscale is adjusted by adaptive gamma transformation to further improve the brightness of the image and reduce the impact of low-light conditions. Second, a LL4PH-Net framework equipped with preset anchors and detection branches is developed to detect low-light traffic objects. The performance of the algorithm is tested on the public BDD100K dataset. Experimental results show that compared with the SOTA methods, the proposed LL4PH-Net acquires the best detection mAP, the most lightweight model and a good time efficiency. It can effectively detect traffic objects even under low-light conditions.