Real-time road lane detection with commodity hardware

Abstract

We present a real-time algorithm for lane boundary estimation based on matched filters and a RANSAC-based method for boundary marking identification. The algorithm is designed to run with minimal assumptions about the environment using a single monocular camera sensor and low-cost commodity compute hardware. With GPU acceleration and TBB, the algorithm runs in less than 25 ms on average on images of size 640 × 480 comparing with 0.12 seconds running on CPU baseline.