Integration of HD Maps and Point Clouds: An Efficient 3D Reconstruction Framework for Autonomous Driving Applications

Abstract. Autonomous driving approaches require simulation environments that accurately converge real-world conditions. These environments must incorporate various factors, including weather conditions, traffic patterns, and unexpected obstacles, to ensure that autonomous systems can effectively learn and adapt. But, most of the frameworks and simulators are using synthetic simulation environments to realize these conditions because of the complexity of representing real-world details and data storage capacity. In these days when autonomous vehicles are close to being included in daily life, this lack of representation could be eliminated by making use of the currently popular 3D reconstruction methodologies that simulate city and road spaces. Their level of detail enhances the training of autonomous systems and helps identifying potential weaknesses in their decision-making processes, ultimately contributing to the advancement of safer autonomous driving technologies. Currently, mapping technologies and geospatial information play a critical role in accurately constituting 3D environments. High-definition maps (HD) are often sufficiently reliable for such tasks because of lane-level representation capability. In this paper, we propose a lightweight 3D synthetic point cloud reconstruction methodology from existing real-world HD maps in the ASAM OpenDRIVE data format by using the Geospatial Data Abstraction Library (GDAL). By leveraging such road network datasets, we aim to improve the efficiency and accessibility of 3D scene reconstruction for autonomous driving applications. Additionally, we aim to provide a low-cost solution to address the annotation bottleneck in point-wise labeling for the computer vision domain with the constructed 3D models.