Road detection for ground vehicle

Road detection for ground vehicle

Unmanned ground vehicles face an overwhelming perception problem: identifying and tracking obstacles while finding the best part of the road. This is even more difficult in offroad conditions. LIDAR based obstacle detection algorithms are in use, but road finding requires using rich visual camera data.

IJK Controls developed an algorithm that detects the road location relative to the unmanned ground vehicle from a front-facing camera. First local orientation of features throughout the image is determined. The image features along a road will generally point in the direction of the road. These orientations are then grouped to find the road in the image. The orientation vectors generally point to a “vanishing point” on the horizon.

By analyzing the structure of the details in the image, the unmanned vehicle essentially follows tire tracks and other road cues just as human drivers do.

keywords: computer vision, recognition, perception, image exploitation, unmanned systems, unmanned ground vehicles