Feature Enhancement Method for Multimodal Nighttime Pedestrian Detection

Nighttime Pedestrian Detection in Low-Light Conditions poses a challenging task, as previous processing methods often introduced noise and limited model generalization due to the low image quality in low-light conditions. This study presents a feature enhancement algorithm tailored for multimodal nighttime pedestrian detection, harnessing the advantages of infrared and visible light data. Our approach utilizes an adaptive weight fusion algorithm and a spatial detail module for multimodal fusion and spatial detail feature extraction, respectively, to accomplish the nighttime pedestrian detection task. Experimental results demonstrate that our method, compared to the baseline algorithm, achieves performance improvements of 3.3%, 11.9%, and 4.5% on the LLVIP dataset.