Thursday 27 March 2025
Pedestrian detection is a critical component of intelligent transportation systems, enabling features like autonomous driving and public transportation monitoring. However, detecting pedestrians in complex environments remains a challenging task, particularly under varying lighting conditions.
Traditional approaches to pedestrian detection rely on single-modal sensors, such as visible-light cameras. These methods are limited by their inability to effectively handle low-light or high-contrast scenarios, where shadows, glare, and other visual artifacts can obscure pedestrian features.
To address these limitations, researchers have explored multimodal sensing, combining data from multiple spectral bands, including visible light, infrared, and thermal imaging. This approach offers a more comprehensive understanding of the environment, as each modality provides unique information about objects and their surroundings.
The proposed PedDet method is an adaptive spectral optimization framework designed specifically for multimodal pedestrian detection. It consists of two core modules: the Multi-scale Spectral Feature Perception Module (MSFPM) and the Illumination Robustness Feature Decoupling Module (IRFDM).
The MSFPM is responsible for fusing features from multiple modalities, adaptively selecting the most relevant information to improve detection performance. This module employs a hierarchical feature extraction strategy, leveraging convolutional neural networks (CNNs) to extract multiscale features from both visible and infrared images.
In contrast, the IRFDM module addresses the challenges of illumination variability by decoupling pedestrian and background features. This is achieved through a novel spatial attention mechanism that selectively emphasizes regions within an image based on their relevance to pedestrian detection. By doing so, PedDet can effectively distinguish pedestrians from background noise even under extreme lighting conditions.
Experimental results demonstrate the effectiveness of PedDet in detecting pedestrians across various environments, including daytime, nighttime, and low-light scenarios. The method outperforms traditional single-modal approaches, achieving a mean average precision (MAP) improvement of 6.6% on the LLVIP dataset, a challenging multimodal benchmark for pedestrian detection.
The visualization of PedDet results showcases the model’s ability to accurately identify pedestrians in complex visual scenes, even under conditions where shadows, glare, and other visual artifacts might obscure features. This capability is made possible by the adaptive fusion of multispectral features and the illumination robustness mechanism.
Pedestrian detection is a critical component of intelligent transportation systems, enabling features like autonomous driving and public transportation monitoring. The proposed PedDet method offers a significant improvement over traditional single-modal approaches, demonstrating its potential to revolutionize pedestrian detection in complex environments.
Cite this article: “Multimodal Pedestrian Detection via Adaptive Spectral Optimization”, The Science Archive, 2025.
Pedestrian Detection, Multimodal Sensing, Adaptive Spectral Optimization, Convolutional Neural Networks, Hierarchical Feature Extraction, Spatial Attention Mechanism, Illumination Robustness, Mean Average Precision, Intelligent Transportation Systems, Autonomous Driving.
