A Comprehensive Study of Object Tracking in Low-Light Environments

TL;DR

The paper tackles object tracking in low-light conditions by studying how distortions such as noise, color imbalance, and low contrast degrade performance and proposing a preprocessing pipeline that combines SUNet denoising and EnlightenGAN enhancement within a transformer-based MixFormer tracker. By training on synthetically dark data and leveraging a Mixed Attention Module (MAM) with an online Score Prediction Module (SPM), the approach achieves superior tracking accuracy over vanilla MixFormer and Siam R-CNN. Key findings show that denoising yields larger gains than enhancement and that training with diverse low-light features improves robustness across varying conditions, with quantified AUC gains when applying the preprocessing at train/test stages. The results have practical impact for surveillance, ethology, and real-time tracking in challenging lighting, and suggest future work toward more diverse distortions and temporal modeling.

Abstract

Accurate object tracking in low-light environments is crucial, particularly in surveillance and ethology applications. However, achieving this is significantly challenging due to the poor quality of captured sequences. Factors such as noise, color imbalance, and low contrast contribute to these challenges. This paper presents a comprehensive study examining the impact of these distortions on automatic object trackers. Additionally, we propose a solution to enhance tracking performance by integrating denoising and low-light enhancement methods into the transformer-based object tracking system. Experimental results show that the proposed tracker, trained with low-light synthetic datasets, outperforms both the vanilla MixFormer and Siam R-CNN.

Figures (8)

• Figure 1: The diagram used for our study on object tracking in low-light scene
• Figure 2: Test results of trackers trained with different noise level. The x axis shows the noise level of the test sets, while the Y axis shows the values of testing metrics.
• Figure 3: Test results of trackers trained with different gamma gains. The x axis shows the gamma value of the test sets, while the Y axis shows the values of testing metrics.
• Figure 4: Images with different levels of gamma gain. Top left shows the image in original daylight environment. Other images has gamma gains of 0.6, 0.3 and 0.2 respectively.
• Figure 5: Test results of trackers trained with different saturation gains. The x axis shows the saturation gain values of the test sets, while the Y axis shows the values of testing metrics.