Project Overview

Visual lane detection and analysis is an increasingly popular area of computer vision research intended to enable computers to automatically determine the position of a road, traffic lane or other similar thoroughfare, within an image acquired by a vehicle-mounted camera.

This project demonstrates one of the more exciting of the numerous applications of this technology - the provision of the primary navigational mechanism for an autonomous robotic vehicle.

The first image (left) above was acquired from the vehicle-mounted camera on the Zoom robotic vehicle (top). The middle image is an edge-detected version of the first. The third image (right) is the source image with some of the results of my Visual Lane Detection and Analysis algorithms superimposed on it. In this particular frame, the vehicle's horizontal position in the roadway is off-center, quite a bit to the left. Also, the vehicle is pointing a bit to the right. The heading correction value generated by my algorithms is 46.9 degrees, which means that in order to get back on track (centered in the lane and pointing forward), the vehicle needs to turn 46.9 degrees to the right of its current heading. This image aquisition - heading correction process is executed more than 10 times per second when controlling a robotic vehicle and it is very effective!

Interested? To learn more about the process, download and view the power point presentation from the downloads section and look at the pictures and videos in the examples section. Computer scientist types are also invited to download the well-documented source code in order learn about how my algorithms are implemented.

My algorithms are self-contained in a Visual Basic module file that runs in both a lab-mode shell and a robot-mode shell. Only code for the lab-mode shell is included on this website - the only real difference in robot mode is that it uses live image capture and dynamically controls a vehicle instead of performing offline processing of previously acquired images.