Geo-Referenced Mobile RGB-D Sensing for Bicycle Lane Width Assessment

Urban bicycle infrastructure requires spatially continuous and quantitative methods for assessing lane width under real operating conditions. This study presents a geo-referenced mobile RGB-D sensing framework for automated bicycle lane width measurement using a lightweight three-wheeled survey platform equipped with a stereo depth camera, GPS receiver, and inertial measurement unit. Bicycle lane boundaries are detected using FastSAM-based segmentation, GPU-accelerated Canny edge detection, and the Hough transform. IMU-assisted depth-to-height conversion distinguishes the road surface from curbs, roadside structures, and obstacles, while temporal-continuity-based lane restoration compensates for faded markings and occlusions. Lane width is estimated from three-dimensional Euclidean distances derived from RGB-D data. The measurements are temporally synchronized with GPS observations and mapped in a geographic information system to visualize spatial width variations along survey routes. Field experiments over four sections totaling 2.87 km showed that 97.79% of the width estimates were within ±5% of the reference width, with a mean absolute error of 0.0218 m. The resulting geo-referenced width maps enable spatially explicit identification of narrow or irregular bicycle lane sections for infrastructure inspection and maintenance.