Road Boundary Detection using Camera and mmwave Radar

Road boundary detection has been an active research area for autonomous driving to support full autonomy in all weather conditions. It also helps human drivers to drive safely in bad weather conditions when vehicles ahead and road boundaries are obscured. For example, knowing the road boundaries enables snow plow vehicles to clean the road more precisely, thereby increasing the amount of drivable area available during the winter. The majority of current road boundary detection techniques use camera and lidar sensors. The camera excels in clear daylight conditions but struggles in low visibility light. While lidar sensors perform well in low light, they struggle in inclement weather conditions such as rain or fog. The high attenuation power of automotive radar makes it extremely effective in all types of weather conditions. However, due to the low resolution of the radar, it is currently limited to object detection for cruise control applications. This paper proposes a method for detecting road boundaries in all weather conditions by combining a camera and mmwave radar. We present radar sensor filters that will aid researchers in making more efficient use of millimeter-wave radars. We demonstrate that our approach performs 20% better than the pure vision-based approach. We showcase that in inclement weather conditions when a camera can barely see our approach can precisely detect road boundaries. The proposed method has been validated by mounting an experimental setup on a test vehicle and driving it in a variety of different conditions and on a variety of different types of roads.