This paper presents a road detection and tracking technique using polarization characteristics of the road in the long-wave infrared (LWIR) spectrum. Conventional vision-based road detection techniques often apply color and texture information, which tend to underperform in low illumination conditions at night. The Division of Focal Plane (DoFP) infrared polarization imaging technology enables real-time acquisition of polarization characteristics of the road with a monocular camera for day-and-night operation. The polarization characteristics of the road in LWIR embody zero-distribution of Angle of Polarization (AoP) in the road region and the difference of Degree of Polarization (DoP) between the road and vehicles. A road detection and tracking scheme is proposed using the difference in polarization characteristics in LWIR between the road region and the background, along with the intensity and temporal information. We also built a LWIR DoFP Dataset of Road Scene (LDDRS) consisting of a total of 2,113 images annotated manually. Experiments on the LDDRS database demonstrate that the proposed method outperforms two state-of-the-art real-time semantic segmentation networks, FANet-34 and SwiftNet, by 1.4% and 2.1% in terms of IoU, respectively.

本文提出了一种利用道路在长波红外 (LWIR) 光谱中的偏振特性的道路检测和跟踪技术。传统的基于视觉的道路检测技术通常应用颜色和纹理信息，这在夜间低照度条件下往往表现不佳。焦平面分割 (DoFP) 红外偏振成像技术能够通过单目摄像头实时获取道路的偏振特性，实现昼夜操作。LWIR中道路的极化特性体现在道路区域极化角（AoP）的零分布以及道路与车辆之间极化度（DoP）的差异。利用道路区域和背景之间长波红外偏振特性的差异，以及强度和时间信息，提出了一种道路检测和跟踪方案。我们还构建了一个 LWIR DoFP 道路场景数据集 (LDDRS)，由手动注释的总共 2,113 张图像组成。LDDRS 数据库上的实验表明，所提出的方法在 IoU 方面分别优于两个最先进的实时语义分割网络 FANet-34 和 SwiftNet，分别高出 1.4% 和 2.1%。