Control Analysis of Road Information Feature Extraction Using Autonomous Vehicle LiDAR

Abstract

Advances in science and technology that makes the autonomous driving becoming a reality The development policy of the autonomous vehicles in Taiwan has begun and the High Definition maps (HD maps) are the key to the success of autonomous driving In order to promote the development of autonomous driving technology and HD maps the Ministry of the Interior of Taiwan has established the High Definition Maps Research Center in 2018 However there are still many difficulties in the construction and maintenance of HD maps To meet the standards of autonomous driving HD maps must possess high accuracy quality and also could be real-time updates Therefore the initial cost of construction is very high so it the maintain The HD maps contain many elements including lane borders traffic signs objects etc which change frequently How to update in time and also offer apply high resolution information under dramatic altering conditions is the goal of many industries and researches In this study we used the low-end LiDAR system on broad the autonomous vehicle to extract and analyze the road information features The scanned point cloud from the Taiwan CAR Lab and then use the normal distribution transform to create a point cloud data layer Then we used the iterative closet point to align point clouds from different scans of the same features Finally we used Canny edge extraction and Poisson surface reconstruction to extract features for 2D road markings and 3D roadside objects and calculated their accuracy and recognition ability The results showed that the proposed method could achieve the accuracy that met the standards of HD maps and also gain a good recognition ability for the traffic signs and nearby objects when the scanning distance is within 5m at an average speed of 15km/h By overlap multi LiDAR data set from various pass of vehicles the data density should be elaborated to fulfill the need of detail mapping Social share point clouds could make the update of HD maps within minutes and offer near real time reality mapping capability

Date of Award	2020
Original language	English
Supervisor	Ting-To Yu (Supervisor)