To address the issue of accuracy degradation in pedestrian detection under low-light conditions, this study proposes a low-light high-precision pedestrian detection algorithm, called LLHPP-YOLO. The main contributions of this study are fourfold. First, an illumination-aware enhancement module is proposed, which leverages the differentiable parameter estimation mechanism of Zero-DCE++ to perform nonlinear illumination correction, thus effectively enhancing feature discriminability in low-light regions. Second, a composite cross-stage feature pyramid architecture, termed CSPSPPELAN, is designed to maintain multi-scale feature fusion capability while reducing the number of parameters in the spatial pyramid pooling layer by 65.5%, so as to improve detection accuracy and frames per second. Third, within the detection head, an efficient reparameterized convolutional attention feature extraction module, termed Rep-EMA, is proposed to enhance the model feature extraction capability. Finally, an optimization strategy for bounding box regression based on the EIoU loss function is introduced, which improves object localization accuracy through an aspect ratio decoupling mechanism. Experimental results on the LLVIP benchmark dataset demonstrate that the proposed model achieves significant improvements compared to the baseline YOLOv7. The detection mean average precision (mAP) increases by 1.4%, computational complexity decreases by approximately 3.1 GFLOPs, the model size is reduced by approximately 5.0M, and the detection speed reaches 400 f/s. Furthermore, LLHPP-YOLO outperforms other mainstream object detection models. Specifically, the mAP is improved by 2.2% and 1.4% compared with that of YOLOv5-L and YOLOv9-C, respectively. The proposed model serves as a high-precision detection solution and can support autonomous driving in low-light environments.