A Comprehensive Accuracy Analysis

Comprehensive Accuracy Analysis of Vehicle Localization with Respect to Imaging Sensor Performance and High-Definition 3D Geospatial Database

PI: Dr. Charles Toth

The availability of accurate and detailed 3D maps of the environment can greatly facilitate advanced driver assistance systems and autonomous vehicle technology in general. High-definition maps contain the complete geometry of the road, location of infrastructure elements, such as traffic signs and lights, at decimeter or even centimeter level accuracy. Since the transportation infrastructure is constantly changing, the challenge is to keep these maps updated. Since more and more sensor data are acquired to support vehicle autonomy, such as for object detection, tracking or obstacle avoidance, etc. This information can be potentially used for mapping purposes. The research question is how these data, acquired by low/modest quality sensors can be jointly exploited to create and, more importantly, continuously update these high-definition maps at the required high accuracy. At the SPIN Lab, we investigate low-cost sensor technologies to create high-definition maps for the automotive industry. Earlier we configured a test vehicle to collect data streams from multiple LiDAR sensor, and still and video cameras. The accurate georeferencing reference for the data acquisition was provided by navigation-grade IMU and GPS. Current research is focused on fusing the data streams, acquired by different sensors in several passes, to create high definition maps of the sensed area.