Article
Infrared and Millimeter-Wave Radar Fusion for Robust Night Pedestrian Recognition
Detecting pedestrians in low-light and nighttime environments is a major challenge for autonomous systems, mainly due to poor visibility and the limitations of relying on a single sensor. To address this issue, this project introduces a multimodal pedestrian detection framework that combines infrared (IR) imaging with millimeter-wave (MMW) radar data to improve detection accuracy, reliability, and real-time performance during night conditions. An enhanced YOLOv5-based deep learning model is utilized to extract meaningful spatial and semantic features from infrared images, while radar signals are processed to obtain important information such as distance, speed, and position of moving objects. To further enhance tracking accuracy, an Extended Kalman Filter (EKF) is applied to reduce noise in radar data and to predict the motion of pedestrians over time. The system performs spatiotemporal fusion by aligning radar target points with corresponding regions in infrared images, followed by a correlation gating mechanism to ensure proper association between the two data sources. Finally, a decision-level fusion strategy integrates the outputs to produce precise and consistent pedestrian detection results. Experimental results show that the proposed approach outperforms single-sensor methods, offering improved precision and robustness under challenging nighttime conditions. This work contributes to enhancing the safety and reliability of autonomous vehicles and intelligent monitoring systems operating in low-visibility environments.
Full Text Attachment
