- About CSAIL
- News + Events
- Alumni & Friends
Albert Huang- Lane Estimation for Autonomous Vehicles using Vision and LIDAR
Autonomous ground vehicles, or self-driving cars, require a high level of situational awareness in order to operate safely and efficiently in real-world conditions. A system that is able to quickly and reliably estimate the roadway and its lanes based upon local sensor data would be a valuable asset both to fully autonomous vehicles as well as driver assistance systems. To be most useful, it must accommodate a variety of roadways, environments with a range of weather and lighting conditions, and highly dynamic scenes with other vehicles and moving objects. Lane estimation can be modeled as a curve estimation problem, where sensor data provides partial and noisy observations of curves. The number of curves to estimate may be initially unknown and many of the observations may be outliers and false detections (e.g., tree shadows or sun-induced lens flare). The challenge is to detect lanes when and where they exist, and to update the lane estimates as new observations are received. This thesis describes algorithms for feature detection and curve estimation, as well as a novel curve representation that permits fast and efficient estimation while detecting and rejecting outliers. Locally observed road paint and curb features are fused together in a lane estimation framework that detects and estimates all nearby travel lanes. The system handles roads with complex geometries and makes no assumptions about the position and orientation of the vehicle with respect to the roadway. Early versions of these algorithms successfully guided a fully autonomous Land Rover LR3 through the 2007 DARPA Urban Challenge, a 90 km urban race course, at speeds up to 40 km/h amidst moving traffic. We evaluate these and subsequent versions with a ground truth dataset containing manually labeled lane geometries for every moment of vehicle travel in two large and diverse datasets that include more than 300,000 images. The results illustrate the advancements made by our most recent algorithms at robust lane estimation in the face of challenging conditions and unknown roadways.