Localization and 3D Mapping using 1D Automotive Radar Sensor

Robin van Gaalen, Faruk Uysal, Alexander Yarovoy
2020 2020 IEEE Radar Conference (RadarConf20)  
The possibility of creating a hybrid lidar and radar based positioning system was vital in creating a robust positioning system. And in this pursuit, it seemed logical to think about how to make the two data sets more relatable to each other, to make them seem, for lack of a better word, more "alike". This spawned the idea of making the created radar map 3D, as the lidar data is also 3 dimensional. Various techniques have been employed to achieve SLAM using radar sensors, either these
more » ... create a two dimensional (2D) mapping using a one dimensional (1D) radar sensor or for three dimensional (3D) mapping, they need a 2D sensor array (for scanning abilities in both azimuth and elevation). The traditional automotive radar being employed today however has a 1D array, that only exploits angular information along the azimuthal angle, and the radar maps are therefore 2D (namely, range and azimuth). To improve situational awareness and localization accuracy, height information about the scatterers (targets) has to be obtained, such that the 2D radar map can be made to be 3 dimensional, look more like the lidar data, and hopefully be easier to relate to the lidar data. Two main approaches exist in literature for estimating the height of the objects by using 1D radar sensors. The first method uses a multi-path approach to exploit the height information by finding the difference in time delay between the line-of-sight (LoS) component of the signal, and the non-lineof-sight NLoS component [10] . The second method makes use of the Doppler signature of targets and is known as Doppler beam sharpening (DBS) [11], [12] . If the radar platform is moving, and the movement of the platform is known with a high enough precision, then the Doppler information for a target, combined with the azimuthal angular information, can be used to deduce the targets height from the ground. In this paper, we address firstly the issue of vehicle localization using a 1D linear radar array in conjuncture with preexisting lidar maps. We demonstrate the possibility to generate a 3D radar map of the environment using a 1D linear array by including the height information acquired with DBS-based processing. Furthermore, some techniques in literature have been tailored or new approaches are proposed specifically for 3D mapping purposes -such as a method for the creation of compatible lidar maps, and a method for estimating the vehicle ego motion-which will be discussed in further sections. Abstract-This paper establishes novel methods for vehicle localization and mapping using a 1D linear automotive radar array in conjuncture with pre-existing lidar maps, and tests if the generated radar map can be made to be 3 dimensional. The basic design of this study was to implement a SLAM (Simultaneous Localization And Mapping) system that co-registers radar data to radar data, and/or register radar data to lidar data. After the execution of experiments, it was established that it is possible to localize the car by relating observed radar data to premade lidar maps, and to continually add to a cumulative map made with the radar data that can further aid the localization process. Furthermore, the radar map created using the 1D linear automotive array can be extended to 3D with proposed processing chain, though more experiments to establish the full potential of this capability are recommended.
doi:10.1109/radarconf2043947.2020.9266487 fatcat:pcr4eukpavehlmtzguevn2stfa