University Archives

A gravel road detection pipeline called StoneScope is presented for identifying forested gravel roads in Advanced Driver Assistance Systems (ADAS) applications. The method focuses on classical computer vision techniques rather than neural networks in order to reduce computational cost and improve explainability. The approach is implemented in MATLAB using standard image processing operations. It combines hue- and texture-based segmentation with morphology-based refinement to isolate the gravel road surface and generate a consistent road mask. Development and evaluation were performed using multiple types of driving video, including scenes with strong shadows, curved roads, and varying lighting conditions. The resulting segmentation produces a gravel road surface mask that can be used for subsequent perception or navigation tasks in rural environments. The work demonstrates that classical image processing methods remain effective for challenging off-road segmentation problems.