<p>In autonomous driving tasks, visual perception systems are gaining increasing attention due to their cost-effectiveness and scalability. However, compared to sensor fusion methods that combine LiDAR and millimeter-wave radar, visual perception often struggles with poor performance in low-light conditions, which significantly impacts the performance and safety of autonomous vehicles. Many existing low-light image enhancement methods fail to effectively improve 3D detection outcomes at night due to camera noise and other factors affecting raw input images. To address this issue, this paper proposes a novel low-light image enhancement method that uses Stable Diffusion as the backbone and incorporates ControlNet to build a multi-condition adapter (SDLN). This method leverages multi-modal inputs, including text, depth, edge, and detail controls, to enhance low-light images. Experimental results on the night scenes of the nuScenes dataset show that the proposed method, referred to as SDLN, significantly improves the visual quality score by enhancing night-time details in images compared to various image enhancement methods. Furthermore, SDLN effectively boosts the performance of advanced 3D detectors for night-time 3D object detection. The BEVDet model achieved a mean Average Precision (mAP) of 17.1%, representing a 4.2% improvement, highlighting its potential for enhancing the safety of autonomous vehicles during night-time driving.</p>