Recently, the use of automated driving technology has been attracting attention as a solution to various social problems related to automobiles. Since the use of digital maps is essential for the realization of automated driving technology, "localization" is an important elemental technology for estimating the position of a vehicle on a digital map. However, existing localization methods require expensive sensors and high-definition three-dimensional map information, which pose a cost problem when applied to ordinary vehicles. In a previous study, a method for localization with a relatively simple sensor configuration was proposed. This method realized localization by using a camera to recognize white lines and road markings, but it could not be used in an environment where there are no white lines, such as a community road. In this study, we developed a method of localization of the lateral direction by detecting the road boundary in a community road. On this point, lateral localization is a technique for estimating the position of a vehicle in lateral direction relative to the direction of travel.
There were two major challenges in the development of the method: the first was the detection of the various road boundaries that exist on the road. The reason for the difficulty is that the existing methods cannot detect all the road boundaries on the road because they are diverse, such as the wall of a residential area or the edge of a parking lot. In this study, we proposed a method for detecting road boundaries by focusing on the gradient change of the road surface as a solution to this problem. The slope change exists at the boundary between a horizontal parking lot and a road, which appears to have no slope change at first glance, due to the slope for water drainage on the road. By using a laser sensor called LIDAR to detect these changes in the cross-slope and height of the road surface, we were able to comprehensively detect the boundary of the road.
The second challenge is to determine the road boundary with a simple sensor. We used a simple laser sensor (LIDAR), but it could only detect partial track boundaries because the amount of information per sensor cycle was small. However, for lateral localization, it is necessary to recognize the road boundary as a pair of both-side lines like white lines. To solve this problem, we developed a method to transform the past recognition results according to the driving trajectory and store them. By accumulating the recognition results, it became possible to detect the left and right road boundaries and estimate the lateral position even with a simple sensor that has little information.
The lateral localization method developed in this study met the accuracy required for autonomous driving on more than 99% of the community roads for which data was collected, demonstrating its practical feasibility. In the future, it will be necessary to expand the system to be able to operate in a variety of road environments, including curved road sections, which were not covered in this study. Although there are still many issues to be solved for the practical use of automated driving technology on community roads, we hope that the results of this research will contribute to the spread of automated driving technology.
An example of a road boundary without white lines in a residential area
Left side: A wall of a house is a boundary of the road. This case has been detected in existing research.
Right side: An example of a road boundary without a three-dimensional structure such as a wall of a house. There is no existing research on this case.
Experimental vehicle. A four-layer LIDAR is installed in front of the vehicle. Unlike the all-round multi-layer LIDAR installed on the top of the vehicle, it is more compatible with existing vehicles.
Left: Observation information from LIDAR mounted in front of the vehicle.
Right: Estimation of the road boundary and the center line based on the time series observation information
Wataru furuse, Takuma Ito, kyoichi Tohriyama, and Minoru Kamata. “Lateral Localization via LIDAR-Based Road Boundary Extraction on Community Roads.” International Journal of Automotive Engineering, doi:10.20485/jsaeijae.11.3_116 .