Revisiting Misalignment in Multispectral Pedestrian Detection: A Language-Driven Approach for Cross-modal Alignment Fusion
This addresses a critical challenge in multispectral pedestrian detection for applications like autonomous vehicles, though it appears incremental as it builds on existing alignment methods with a new semantic approach.
The paper tackled the problem of misalignment between RGB and thermal modalities in multispectral pedestrian detection, especially under real-world conditions with heavy misalignment, by introducing a language-driven framework that uses Large-scale Vision-Language Models for cross-modal semantic alignment to enhance detection accuracy without costly pre-processing.
Multispectral pedestrian detection is a crucial component in various critical applications. However, a significant challenge arises due to the misalignment between these modalities, particularly under real-world conditions where data often appear heavily misaligned. Conventional methods developed on well-aligned or minimally misaligned datasets fail to address these discrepancies adequately. This paper introduces a new framework for multispectral pedestrian detection designed specifically to handle heavily misaligned datasets without the need for costly and complex traditional pre-processing calibration. By leveraging Large-scale Vision-Language Models (LVLM) for cross-modal semantic alignment, our approach seeks to enhance detection accuracy by aligning semantic information across the RGB and thermal domains. This method not only simplifies the operational requirements but also extends the practical usability of multispectral detection technologies in practical applications.