Page 59 - Read Online

Huang et al. Complex Eng Syst 2023;3:2 I http://dx.doi.org/10.20517/ces.2022.43 Page 3 of 20

tems (ASAM), has been used in simulations for some time and has good landing performance in some as-
sisted driving models. The Navigation Data Standard (NDS) is a standard format for vehicle-level navigation
databases jointly developed and published by vehicle manufacturers and automotive suppliers. The NDS for-
mat enables sharing of navigation data between different systems by separating the navigation data from the
navigation software. Although OpenDRIVE and NDS are formats developed by more authoritative organiza-
tions, they need to be more open, as they provide only partial information to most developers. Therefore it is
challenging to use them in practice [18,19] .

ApolloOpenDRIVEisamodifiedversionofOpenDRIVEtoaccompanytheBaiduApolloautonomousdriving
system. Insteadofusinggeometricelements,itusessequencesofpointstorepresentroadelements. Inaddition,
Apollo OpenDRIVE stores reference lines on the map and then describes the lane lines relative to the reference
lines. This allows Apollo OpenDRIVE to express maps with higher accuracy than OpenDRIVE for the same
map file size and also facilitates some calculations in the subsequent planning module.

In 2018, Poggenhans et al. [20] released the open source Lanelet2. Based on the OpenStreetMap (OSM) format,
Lanelet2 has been extended and allowed direct access to many of the open source tools that accompany OSM.
Benefiting from its complete toolkit, open architecture, and easily editable features, Lanelet2 not only allows
the storage of information about roads, road signs, light poles, and buildings with precise geometry but also
enables lane level and traffic-compliant routing.

2.3. Multisensor fusion localization
GNSS are widely spread in intelligent transport systems and offer a low-cost, continuous and global solution
for positioning [21] . It can provide a more stable location. GNSS localization system has obvious disadvantages:
significant errors and easy to be obscured. Therefore, scholars have increasingly recognized multisensor fusion
as necessary in recent years. Simultaneous localization and mapping (SLAM) is a technology that constantly
builds and updates environmental information by sensing things in an unknown environment while tracking
their position in the background. SLAM is generally divided into light detection and ranging (LiDAR)-based
SLAM, such as LOAM [22] , LeGO-LOAM [23] , LINS [24] , and LIO-SAM [25] , and vision-based SLAM, such as
ORB-SLAM [26] , VINS [27,28] . If the localization relies solely on LiDAR or cameras, position estimation errors
will accumulate over a long time and distance. An HD map, as a globally consistent data source, can also
provide reliable global location constraints. Multisensor fusion localization algorithms combined with HD
map lane-level localization algorithms will be more accurate and have great potential.

2.4. Localization based on HD map
Scholars continue to reduce the error by matching pavement marking features or lane line curvature based
on existing localization [6–12] . However, the visibility of road markings is affected by light. The visibility of
different markings on the same road segment varies greatly at different periods, making it difficult to achieve
stable positioning performance. At the same time, these efforts do not consider the common function of
road elements in localization and planning, and these methods can only use the generated road elements in
localization.

3. SYSTEM OVERVIEW
The proposed system consists of three parts. The first part is lane line fitting. An inverse lane line perspective
mapping method combined with ground equations is discussed. An iterative approximation-based process
of fitting piecewise CHS curves is proposed. This method satisfies the requirement of small data storage and
ensures the continuity of lanes.

The second part is the HD map postprocessing. The data structure and coordinate system required for the HD