Smalley, Chris, Hong Nguyen, Mark Paulik, Mohammad Utayba, Mohan Krishnan, and Chaomin Luo
A novel image processing algorithm is introduced to address the roadway lane identification task associated with self-driving or autonomous vehicles. Images are captured with a monocular, color, high-resolution camera in RGB format. Initially, to reduce specular noise and localized glare, a real-time binning operation is applied to average 2x2 image sub-regions. This is followed direct color plane manipulation to enhance contrast and produce a gray scale image for subsequent analysis. Adaptive thresholding, binary morphology, and connected component analysis are then employed to produce a clean binary image that effectively highlights roadway lanes. These techniques could be interchanged depending on the conditions and the algorithm can be tailored to fit specific course attributes.