The YARF system for vision-based road following

The YARF system extracts road features from color images in order to drive a vehicle along the road. Central to the system is a model of the road which includes both geometric information about the relative placement of the features defining the lane structure of the road and information about the appearance of those features. This model is used to selectively apply specialized image segmentation methods to perform the feature detection; to provide the system of constraints which relate the feature positions to the vehicle position on the road; and to provide a context of expectations about the appearance of the road which can be used to analyze situations where expected features are not seen. Experimental results are presented showing the benefit of using multiple specialized image segmentation techniques, the advantages of using Least Median Squares estimation in situations where there are outliers in the data, and the ability of YARF to detect intersections and changes in lane structure based on the failure to see expected features in the image.