No abstract

No abstract

No abstract

The revised unpaved road detection system (RURD) is a novel method for detecting unpaved roads in an arid environment from color imagery collected by a forward-looking camera mounted on a moving platform. The objective is to develop and validate a novel system with the ability to detect an unpaved road at a look-ahead distance up to 40 meters that does not utilize an expensive sensor, i.e., LIDAR but instead a low-cost color camera sensor. The RURD system is composed of two stages, the road region estimation (RRE) and the road model formation (RMF). The RRE stage classifies the image patches selected at 20-meter distance from the camera and labels them to either road or non-road. The classification result is used as a high confidence road area in the image, which is used in the RMF stage. The RMF stage uses log Gabor filter bank to extract road pixels that connect to the high confidence road region and generates a 3rd degree polynomial curve to represent the road model in a given image. The road model allows the system to extend the detection range from 20 meters to farther look-ahead distance. The RURD system is evaluated with two-years worth of data collection that measures both spatial and temporal precisions. The system is also benchmarked against an algorithm from Rasmussen entitled "Grouping Dominant Orientations for Ill-Structured Roads Following", which shown an average increase detection accuracy over 30 [percent].

In this paper, we propose a reinforcement random forest algorithm as a novel approach to detect unpaved road regions at stand-off distances. A random forest classifier is used to differentiate between road and non-road pixels/patches without over fitting the training data. Utilizing a reinforcement technique, the algorithm can handle foreign objects that we encounter in real world driving. Furthermore, classifying road patches at different distances generates multiple levels of road agreement for each pixel within the image. Using different threshold values of this agreement level provides adaptability to the road finding results. The selection of low threshold values produces better detection rates but also increases false alarms. On the other hand, high threshold values lower the detection rate and decreases false detections. In our experiments, the proposed algorithm is tested on color video of unpaved road in an arid environment.