Low Cost Vision-Based Real-Time Lane Recognition and Lateral Pose Estimation

Real-time road lane recognition and position estimation algorithm in small vehicle is generally limited by the amount of processing power available. The goal of this research is to develop a light-weight algorithm to recognize a pair of road lane markers using computer vision and to estimate the vehicle position relative to the lane markers. The road lane recognition algorithm uses inverse perspective mapping of detected lines to speed-up recognition of the lane markers pair, and then a color matching stage is employed to reduce false recognition. The algorithm is implemented in a small battery-powered processing unit and evaluated in a miniature four-wheel vehicle to recognize a pair of lane markers on the floor and to estimate the vehicle's pose on the floor relative to the lane markers. The algorithm managed to estimate the lateral position and the orientation of the vehicle with accuracy about 1.5 cm and 2 degree respectively, and an estimation rate of 2.7 Hz.