Sophisticated digital map is an essential resource for intelligent vehicles to localize and retrieve environment information. However, the open map resource do not contain enough information for decision making during autonomous driving. Although comprehensive commercial map can provide precise map knowledge, the data format is not in a machine-readable format. Therefore, we retrieve useful knowledge from high-precision commercial map and convert it into ontology based data to help intelligent vehicles perceive driving environment and make decisions at various tra c scenarios. Furtheremore, the converted map data can be used as a golden standard for evaluating tra c sign detection, road mark detection, and automatic map construction.
Developing intelligent or autonomous vehicles is bene cial to the society not
only for assisting disabled or elderly people, but also for improving safety and
transportation e ciency. One of the most challenging problems is to enable
intelligent vehicles to drive safely by perceiving driving environments. In order
to make intelligent vehicles think as human drivers do, we have to represent
environment information in a machine-understandable format. Ontology can be
used to represent environment information in a machine-readable format for
intelligent vehicles to make safety decisions [
Without sophisticated map information, it is impossible to make intelligent vehicles drive safely on urban roads. Current public map data lack detailed map information such as lane information and road signs. Although, precise commercial map data can provide high-precision map information, it is di cult to access knowledge for performing reasoning because of its speci c data format. In this paper, we introduce ADAS ontology for autonomous driving tasks and ADAS ontology-based map data, which is automatically constructed by retrieving knowledge from comprehensive commercial map data. (a) Classes.
(b) Object properties. (c) Data properties.
The TTI core map ontology5 was introduced to develop a decision making system
for intelligent vehicles [
The ADAS ontology contains 93 classes, 21 object properties, and 39 data properties. As shown in Fig. 1a, we added various types of lanes such as AcceleratingLane, DeceleratingLane, LaneExJunction (lanes that exit a junction), and LaneInJunction (branching lane and merging lane inside a junction). Various types of lanes can help intelligent vehicles perceive environment at junctions and choose a proper lane to run safely. StopLine and Tra cSignArrow are also added as the subclass of InstructionSurfaceSign and Tra cSignal, respectively.
Additional object properties (Fig. 1b) and data properties (Fig. 1c) are also added to assign semantic knowledge of map. \entryLane" and \exitLane" are speci ed for lane connections and \relatedTra cSign" is used to describe relations between lanes and tra c signs. We added extra data properties (allowLaneChangeLeft and allowLaneChangeRight) to assist lane changing and added identi cation codes, coordinates, and ids for tra c signs and tra c lights. 5 http://www.toyota-ti.ac.jp/Lab/Denshi/COIN/Ontology/TTICore-0.1/ 6 http://www.toyota-ti.ac.jp/Lab/Denshi/COIN/Map# 7 http://www.semanticweb.org/example/ontologies/ADAS00#
An Ontology based Map Converter for Intelligent Vehicles (a) Map data area. c 2016 Geospatial In- (b) A schematic illustration of the high formation Authority of Japan (GSI). precision 3D map for automated driving. The commercial map data covers the road networks (blue dot line) shown in Fig. 2a, which is around Wakamatsu Campus of Kyushu Institute of Technology in Japan. It's a part of ZENRIN high-precision map designed in the aim of automated driving and advanced driving assistant systems, provided by ZENRIN Co., Ltd.. The high-precision 3D map data contains lane-level information and precise positions of objects such as tra c signs and tra c lights. Fig. 2b visualizes the 3D high-precision commercial map data (an intersection in red rectangle in Fig. 2a), which we converted for developing decision making systems.
JSON (JavaScript Object Notation) is used to convert the commercial data
format into ontology format, which is a text format that facilitates structured
data interchange between all programming languages [
The map converter implemented using Python converts commercial map data into ADAS ontology-based map data automatically. The converter outputs data in Turtle (Terse RDF Triple Language) format as shown in Table 2. The converted data speci es types of lanes and precise coordinates of enter and exit points of each lane. Lanes in a junction are connected with lanes outside of a junction using the property \adas:exitLane" and \adas:enterLane".
This map converter retrieves essential parts of the features in commercial map data, that are useful for intelligent vehicles to perceive environment information. This ADAS ontology-based map data will be used to develop collision avoidance, decision making, or adaptive control systems for autonomous vehicles. 4
We introduced ADAS ontology, which was used to convert precise commercial map data into ontology-based data. JSON was used to interchange commercial map format to ADAS ontology concepts. We will use the ADAS ontology-based precise map data to develop decision making systems, which will use the knowledge of lane connections at junctions, tra c signals, and tra c signs.
This work was supported in part by the New Energy and Industrial Technology Development Organization (NEDO) \Next-generation Arti cial Intelligence".