LARGE-SCALE IMAGE REGISTRATION FOR ROAD MARKING DETERIORATION MANAGEMENT FROM IN-VEHICLE CAMERA IMAGES AND LOGGED CAN DATA

摘要

There are many studies to detect and recognize road markings from in-vehicle camera images. It is necessary for autonomous driving to recognize road markings by using in-vehicle cameras, but the recognition becomes difficult when the road markings have been deteriorated. Therefore, we thought of managing road markings by measuring conditions and by acquiring data continuously. As a preliminary stage to realize the management of road markings, large-scale road image registration was implemented. It requires information obtained from the common vehicle equipment such as in-vehicle camera and Control Area Network (CAN). Similarity based on ZNCC (Zero-mean Normalized Cross-Correlation) is calculated for pattern matching, then continuous images are overlaid through image registration. We also proved effectiveness by evaluating the accumulated error of yaw angle between the estimated value and the ground truth.