Chapter 24 Recognition of Road Surface Condition Through an On-Vehicle Camera Using Multiple Classifiers

摘要

This paper describes development of a system that identifies the condition of a road ahead, using a mono-camera mounted behind the windscreen of a BMW autonomous driving research prototype. Multiple regions of interest are extracted from the camera image and processed with various classifiers to obtain a valid result. The regions of interest include the wheels of cars driving in front to observe spray water or dust, the road behind a car driving in front to observe reflections, and a bird view transformed snippet of the road surface identify various conditions (dry surface, wet surface, snow, ice, dirt, cobblestone, gravel and asphalt) and additional features (potholes, cracks and leaves). The goal of this system is to enhance safety when driving autonomously under adverse weather or road conditions. The data retrieved by this system can be used to estimate the friction of the road surface. This is a work in progress paper. The prototype in the current state of development can classify wet and dry roads with an accuracy of 86%.