Thesis Presentation

Previous lane-line trackers, such as lane departure warning systems, have focused on the current lane only, ignoring the lines of other lanes and road curbs. The goal of the developed system is to detect and track all the lane curves in the road scene, including exit lanes (target split), on-ramps (target merges), dashed lines, and other pavement markings. To do this, each input image is first transformed to a top-down view using the cameras projection parameters, in which high-intensity curve segments with certain widths are located by applying a set of filters. Then by using telemetry data of the car motion, the top-down view images are combined to a mosaic image corresponding to the road surface that the car traveled most recently. Line segments in the mosaic are grouped into individual lane curves, and their equations are estimated. This process is repeated as the car moves forward.

Thesis Committee:
Takeo Kanade
Simon Baker