Computer Vision System for Road Surface Marking Recognition

The paper proposes a prototype of embedded system for road surface marking recognition. An overview of the most relevant approaches for road surface marking detection has been provided. The approach used in the paper is based on the machine learning method. According to the review using neural networks allows to achieve good accuracy and high generalizability. A single board computer was used as a hardware part of the proposed system. The system provides recognition of road surface markings with an accuracy of up to 91% at a distance of up to 36 meters. A single frame processing time is 3.75 seconds. Methods for increasing the performance of the developed road surface marking recognition system have been proposed.