Urban Road Detection For Unmanned Ground Vehicles Based On Binocular Visual System

Robust and real-time lane detection is an important research topic of unmanned ground vehicles navigation. As such, it has been an active field of research with considerable progress made in the past few years. For example, a lot of mature algorithms for lane detection have been developed for highways. But lane detection is still an open problem for the sake of various scenarios and the diversity of lane structure.In this paper we give an emphasis on the lane detection in urban streets and present a structural lane detection method based on binocular visual system. We use two cameras simultaneously to achieve enough environment information around the unmanned ground vehicles. First of all, we define the lower half of the image as region of interest to avoid the disturbance from pedestrians and other dynamic obstacles. And we generate this region into the "bird’s view" by using Inverse Perspective Mapping based on the calibration information of the camera, which will benefit for the lane detection since the lanes that appear to converge at the horizon in the origin image turn to be parallel in the IPM image instead. We also filter the image with two dimensional Gaussian kernel to reduce the influence of the environment, which is followed by a progress of threshold. Second, we extract the beeline features through Hough transform and calculate weights for each of them. As the serial images are analyzed, we can give a reasonable prediction of the lanes’locations according to the results from the previous frames and thus eliminate the beelines with considerable bias. At last, we propose a lateral model with the lane width to find the optimal lanes and merge the results from two cameras.We conduct our experiments on the structural road in the campus and run the algorithm both on the on-line mode and off-line mode. From the analysis of the experiment results we can conclude that the precision and real-time performance is good enough for the navigation of the unmanned ground vehicles.