Radar Based Estimation of Ditches in the Vicinity of the Road

Abstract: Radars used for detecting objects with a positive elevation, such as other vehicles, are common in autonomous parking and braking systems in modern vehicles. Detecting objects with a negative elevation, such as ditches and holes, is however more troublesome. A common approach is to use a lidar, but a lidar is very costly and fragile compared to a radar. In this thesis, two two-dimensional radars are attached above the windshield of a truck and aimed down towards the ground. At first the geometrical limitations of detecting ditches is analyzed in order to find which mounting angles of the radars are viable. Data is then collected from the radars, with the determined angles, a Global Positioning System (GPS) unit, and Inertial Measurement Unit (IMU) by driving the truck in a real world terrain. Data from a lidar is also recorded for reference. Combined with a GPS and IMU, the radar detections are first transformed from the radar coordinate system, the a truck coordinate system, and finally to global Universal Transverse Mercator (UTM) coordinates. The global position of each detection is filtered, and finally used to create an elevation map of the environment. A similar map is also created from the lidar detections. The resulting radar elevation map accurately maps the terrain near the vehicle, including the ditches next to the road. The radars appear to miss small objects, and the density of the detections is quite low for the radar mount angles used. In order to improve the accuracy, the vertical position of the radar detection needs to be determined. A higher density of detections would also improve the mapping, which could be aquired by decreasing the pitch angle of the radars.