High-definition maps of road environments
Project Leader: Kourosh Khoshelham
Primary Contact: Kourosh Khoshelham (firstname.lastname@example.org)
Keywords: computer vision; deep learning; geomatics; machine learning; spatial computing
Disciplines: Infrastructure Engineering
An essential component of a fully autonomous vehicle is a specialised highly detailed map, usually referred to as high definition (HD) map, which contains the 3D location of all traffic signs, traffic lights, trees, and every relevant object in the road environment. Efficient generation of HD maps remains a major challenge for the automotive industry.
The aim of this project is to develop methods for accurate and domain-adaptive object recognition in point clouds for efficient generation of HD maps that enable autonomous navigation. The hypothesis to be tested in this research is that accurate and domain-adaptive 3D object recognition can be achieved by training deep convolutional networks using large numbers of synthetically generated training samples across different domains.