In recent years, research on intelligent vehicles has been conducted for various purposes. Although various forms of intelligent vehicles, such as autonomous driving and advanced driver assistance systems, have been envisioned depending on the scenario of actual operation, intelligent vehicles need to recognize the driving environment as well as human drivers. However, there are limitations in recognizing the driving environment using on-board sensors, and there are also issues in terms of cost when installing overly sophisticated sensors. Therefore, the idea of using the digital map as a pseudo-long-range sensor has emerged, considering the information on the digital map as equivalent to the sensing results of the onboard sensor. For example, the idea that recognizing the existence of a stop line in front of a vehicle by an onboard sensor is equivalent to knowing the existence of a stop line in front of a vehicle by a digital map. Based on this idea, digital maps have come to play an important role in the research of intelligent vehicles.
In considering digital maps for mobility, the following three points are major issues.
What kind of data structure is desirable for digital maps from the viewpoint of vehicle control?
How do we create the contents of the digital map?
How does the vehicle localize on the digital map?
In the field of autonomous mobile robots, a method for simultaneously localizing and creating digital maps called SLAM has been proposed, and it has also been proposed to be used in automobiles. However, unlike autonomous mobile robots, it is difficult to use such methods for owner-operated cars that do not have a limited driving area, although it is not a problem for service cars that have a limited driving range. Therefore, our research group has proposed a new digital map called "LeanMAP" for autonomous driving by extending the digital map proposed in the mid-2000s. Specifically, we proposed the data structure of the digital map from the viewpoint of on-board map data and vehicle control. We also proposed a method to deepen the map data of the car navigation level to the level of a semi-high-precision map for autonomous driving by using a data-driven method based on the driving data of actual vehicles in order to reduce the resources for constructing map data. Further, to digital such electronic maps, we proposed a self-positioning method based on landmark detection by on-board sensors and matching with map data.
Our new digital map not only has the advantage of low data volume for in-vehicle use, but also has the advantage of lowering the requirements for the sensors installed in the vehicles that use it. It also has the advantage of relatively low resources for map data maintenance. Therefore, it is expected to contribute to the realization of automatic driving and advanced driver assistance systems in owner-operated cars for public roads as a technology that exists as an extension of current automobile development.
Takuma Ito, Masahiro Mio, Kyoichi Tohriyama, and Minoru Kamata, "Novel Map Platform based on Primitive Elements of Traffic Environments for Automated Driving Technologies", International Journal of Automotive Engineering, 7(4), pp. 143-151, doi:10.20485/jsaeijae.7.4_143.
Takuma Ito, Satoshi Nakamura, Kyoichi Tohriyama, and Minoru Kamata, "Data-based Modification System of LeanMAP Contents for Automated Driving", International Journal of Automotive Engineering, 9(3), pp. 115-123, doi:10.20485/jsaeijae.9.3_115.
Takuma Ito, Satoshi Nakamura, Kyoichi Tohriyama, and Minoru Kamata, "Deepening method for LeanMAP content based on a virtual trajectory by lateral transcription", Mechanical Engineering Journal, 6(3), doi:10.1299/mej.18-00558.
Methods for localization on a digital map can be roughly classified according to the data format of the digital map. For example, in the SLAM and VSLAM methods, the digital map is constructed based on the feature points observed by the sensors, and the method of matching the feature points described in the map with those observed in real time is adopted. On the other hand, in our proposed LeanMAP, the map is described by standardized driving environment components, so that the localization is conducted by matching the landmarks described in the map with those detected by the sensors. Therefore, the technologies to detect landmarks by on-board sensors such as cameras and LIDAR are necessary.
In our research group, we aim to realize autonomous driving and advanced driver assistance systems on public roads with equipment at the level of private cars in the near future. Therefore, landmarks for localization need to exist on the public road and need to be encountered with a reasonable frequency. The landmarks to be detected include stop lines, crosswalks, speed limit signs, and diamond-shaped crosswalk warning signs. Additionally, in contrast to urban areas, rural areas do not have sufficient budget for road maintenance, and some of these landmarks may have been lost due to wear and tear. Therefore, our research group has developed a new detection technique that can detect even grazed road markings by focusing on the combination of partial features of the road markings, making good use of map information.
Detection of damaged stop line and crosswalk
Satoshi Nakamura, Takuma Ito, Toshiki Kinoshita, and Minoru Kamata, "Detection Technology of Road Marks Utilizing Combination of Partial Templates", International Journal of Automotive Engineering, 9(3), pp.105-114, doi:10.20485/jsaeijae.9.3_105.
Takuma Ito, Kyoichi Tohriyama, and Minoru Kamata, "Detection of Damaged Road Paints of Crosswalks by Focusing on Multi-layered Features", International Journal of Automotive Engineering, 10(4), pp. 356-364, doi: 10.20485/jsaeijae.10.4_356.
Takuma Ito, Kyoichi Tohriyama, and Minoru Kamata, "Detection of Damaged Stop Lines on Public Roads by Focusing on Piece Distribution of Paired Edges", International Journal of Intelligent Transportation Systems Research, doi: 10.1007/s13177-020-00220-7.