<p>In recent years, with the rapid development of information intelligence and network technologies, multi-object tracking (MOT) has been widely adopted in practical scenarios. However, MOT in low-light environments (e.g., nighttime or rainy conditions) often suffers from insufficient brightness, low contrast, and color distortion, leading to difficulties with target feature extraction and reduced reliability of detections, thereby degrading tracking performance. To address these issues, this paper investigates the problem of MOT in low-light scenarios. First, guided by noise source analysis, we design a two-stage denoising low-light image enhancement algorithm that effectively enhances image feature quality. Second, building on this, we propose a low-light MOT algorithm named LE-Track, which integrates two-stage denoising combined with feature enhancement and association. LE-Track employs a two-stage denoising enhancer to improve image features, significantly improving detection accuracy and feature discrimination while ensuring detection reliability and suppressing background interference. Furthermore, the algorithm optimizes similarity metric by combining Mahalanobis distance and shape similarity, and adopts a two-stage association strategy for handling target appearance variations and dense occlusions. Experimental results demonstrate that our method achieves superior performance on the MOT17 benchmark, the synthetic LLMOTD, and a real-world low-light dataset LMOT_real, thus validating its effectiveness and robustness in low-light scenarios.</p>