1 Introduction

The development of intelligent transport systems has resulted in autonomous vehicles that can be used for travel. This development is expected to reduce annual traffic accidents by millions and relieve traffic congestion. Autonomous vehicles rely on sensors, such as positioning sensors, light detecting and ranging (known as LiDAR), and cameras, to enable real-time computing and decision-making. Producing sensors with high precision and computing efficiency are the two focuses of sensor research and development. High-definition maps (HD maps) play a key role in sensor development.

HD maps and conventional electronic navigation maps differ in their geometric accuracy and attribute content. The evolution of car navigation maps is summarized in Fig. 1 (Chiang et al., 2019). Electronic navigation maps provide users with navigation services, such as road information and route planning; their horizontal accuracy is approximately 5–10 m. However, autonomous vehicles require maps with higher precision to avoid collisions. Self-driving applications require cm-accurate, in the area of 10–20 cm, active control and localization (Aeberhard et al., 2015). TomTom (tomtom.com) released HD maps firstly in 2015, with coverage across Europe, America and Asia. Such HD maps features lane geometry with 15–20 relative accuracy [3]–(TomTom, 2020a, b, c). HERE (here.com) offers HD maps with multiple layers, which includes road model, lane model, and localization model with 20 relative accuracy (Liu et al., 2020). As for HD maps in Asian countries such as Japan and Korea require horizontal and vertical accuracy of 25 and 35 cm, respectively (Chiang et al., 2022). In Taiwan, HD maps require horizontal and vertical accuracy of 20 and 30 cm, respectively (HD maps operation guideline v2, 2019).

Fig. 1
figure 1

Evolution of car navigation maps

Full size image

The Society of Automotive Engineers International (SAE) has classified autonomous vehicle systems into six levels (Levels 0–5) (SAE International, 2021). Currently commercially available vehicles with Advanced Driver Assistance Systems are Level 2; fully autonomous vehicles without human control are Level 4 or greater. Level 4 self-driving requires high navigation accuracy to maximize safety; thus, HD maps are required for the development of self-driving technology. HD maps are helpful for all levels of autonomous vehicle, but become increasingly critical as the level of automation is increased (Stephen & Michael, 2020). The map accuracy required for each self-driving level is listed in Fig. 2 (Shen et al., 2019).

Fig. 2
figure 2

Map accuracy requirements of various self-driving levels

Full size image

In addition to providing auxiliary information that increases autonomous vehicle safety and facilitates route planning, prior observations in HD maps can be correlated with the data acquired by sensors mounted on autonomous vehicles; that is, maps can assist the vehicle in identifying objects or traffic signs. Therefore, the quality of HD maps affects the performance of self-driving technology. HD maps produced by surveying companies must be verified to meet the requirements of autonomous vehicles.

According to the ISO 18750 standard for Local Dynamic Maps (LDM), HD maps are considered permanent static maps (ISO 18750, 2018), with the goals of assisting in accurate navigation and providing accurate lane-level road network information for mobile decision-making systems. Fig. 3.

Fig. 3
figure 3

LDM structure (Shimada et al., 2015)

Full size image

To enhance the development of commercial surveying and self-driving technology, a unified national HD maps standard must be established. Mapping and self-driving technology developed on the basis of permanent static maps can be used to establish a static environment simulation for driving and testing. In 2019, the Ministry of the Interior of Taiwan (MOI) published several standards and guidelines for standardizing the production, validation, and verification procedures for HD maps. An overview of the MOI standards and guidelines for HD maps production is displayed in Fig. 4 (Shen et al., 2019).

Fig. 4
figure 4

MOI recommendations for HD map production

Full size image

The HD map supply chain in Taiwan comprises static map data collection through Mobile Mapping Systems (MMSs), vector map data production, and validation and verification of HD maps. The HD map supply chain architecture is presented in the dotted box in Fig. 5 (Verification and Validation Guideline for HD Maps, 2020).

Fig. 5
figure 5

HD map supply chain architecture

Full size image

The static base map comprises point clouds (LAS format), a vector map (SHP files), and an OpenDRIVE format. The standardization of the base map in Taiwan is shown in Fig. 6. In this paper, procedures for validating and verifying the accuracy of HD maps are proposed. The procedures are divided into six steps: operation planning verification, control survey verification, operation result verification, point cloud validation, vector map validation, and OpenDRIVE simulation.

Fig. 6
figure 6

Standardization of the base map in Taiwan

Full size image

2 Verification and validation procedure

In this section, the method to implement the verification and validation procedure is introduced. The content in previous procedure should be qualified to get into the next procedure. If there are any content is unqualified, the content should be revised otherwise the HD map procedure cannot be proceeded. The flowchart of the verification and validation procedure is shown in Fig. 7.

Fig. 7
figure 7

Flowchart of the verification and validation procedure

Full size image

2.1 Operation planning verification

Operation planning is the foundation of surveying and mapping. The safety of the surveying process and the environment of the surveying area must be evaluated before survey route planning is performed. The characteristics of each survey area should be considered during the verification process to enable the implementation of the subsequent survey procedures.

2.2 Control survey verification

The goal of the control survey is to acquire the true value of control points and check points. Control points are points that are used to adjust observed data, reducing errors and residuals. The number of the control points depends on the performance of the Inertial Navigation System (INS) equipped on the MMS which is used for the HD map procedure. The recommended amount of the ground control points related to the INS specification are shown in Table 1 (HD maps operation guideline v2, 2019). Check points are used for accuracy analysis and are excluded from the adjustment process. There are at least 10 check points shall be set in each survey area. If a surveying area is more than 5 km, two check points shall be added for every 1 km extra. A total of 10% of check points and control points are sampled in the proposed verification process. The horizontal and vertical accuracy requirements for control points and check points are 10 cm and 15 cm, respectively. The control points and check points are typically selected to be obvious features, such as corners of road markings or manhole covers. A photograph of the control survey verification process being performed is displayed in Fig. 8.

Table 1 Recommended amount of the ground control points
Full size table
Fig. 8
figure 8

Control survey verification

Full size image

2.3 Operation result verification

After completing the survey, the raw point cloud and image data acquired using MMS should be verified before postprocessing to ensure that the data meet relevant requirements. The completeness and the correctness of raw data, including inertial navigation system, global navigation satellite system, odometer, trajectory, and raw image data, are verified in this step. An example raw point cloud and corresponding image are displayed in Fig. 9.

Fig. 9
figure 9

Raw point cloud and corresponding image

Full size image

2.4 Point cloud validation

The point cloud data are saved in the LAS format; geometric coordinate, intensity, scan line, and GPS time data are required. Point cloud data that are used as the constraint data for a self-driving system must be classified as the ground and nonground points to avoid a car control system making a misjudgment due to noise occurring in an environment with occlusion. The point density requirements are divided into three levels on the basis of the positioning accuracy (Table 2) (Verification and Validation Guideline for HD Maps, 2020). The road edge plus 1 m is defined as the border of the point density validation area, and 1-square-meter point clouds are sampled for validation. In each survey area, unqualified samples should be less than 5% of the all samples. If the point cloud data fail validation, it can be further corrected or rescanned if the data are unsalvageable.

Table 2 Point density level required for various applications
Full size table

To achieve the point density requirement, MMS usually scans each survey area several times to reduce the occluded area. However, errors occur in each scan line; thus, adjustment is required. In the point cloud adjustment process, ground control points are used to increase the relative accuracy of scan lines. Absolute accuracy can also be increased by using the control points, resulting in the point cloud quality meeting the HD maps requirements. A point cloud before and after adjustment is presented in Fig. 10.

Fig. 10
figure 10

Comparison of a point cloud before and after adjustment

Full size image

The relative accuracy validation of the point cloud ensures that system errors between each scan line are corrected during point cloud adjustment process. This validation has two steps: relative elevation and relative horizontal accuracy validation. First, the relative elevation accuracy is validated; a validation position is selected for every 100 m in the survey area. A 5 × 5 m2 area centered on this position is then selected, and the point cloud in this area is removed. The elevation of this plane for each scan line can then be determined. The relative elevation accuracy is the elevation difference between the maximum and minimum elevations of a set of overlapping scan lines and should be less than 10 cm. Relative elevation accuracy validation is presented in Fig. 11. The second validation step is relative horizontal accuracy validation, which can be performed by examining traffic signs, poles, or islands in the survey area. For example, the center of a traffic sign in each scan line can be used as a point for calculating the horizontal error. The relative horizontal accuracy is the difference between the maximum and minimum horizontal values of this point for a set of overlapping scan lines and should also be less than 10 cm. A depiction of the relative horizontal accuracy validation is presented in Fig. 12.

Fig. 11
figure 11

Relative elevation accuracy validation

Full size image
Fig. 12
figure 12

Relative horizontal accuracy validation

Full size image

2.5 Vector map (SHP file) validation

A vector version of the HD maps can provide accurate geometric coordinates and attribute data for assisting autonomous vehicles in navigation and decision-making; these data include lane lines, lane curvature, lane markings, and traffic sign and light locations. The geometric accuracy and the attribute data of the vector map must also be validated. The geometric accuracy is verified by calculating the absolute and relative accuracy. The absolute accuracy must be 20 cm in the horizontal and 30 cm in the vertical directions; relative accuracy must be 10 cm in the horizontal and 15 cm in the vertical directions.

According to the HD Map Data Content and Formats Standard published by the MOI, HD maps content should include roads, lanes, traffic signs, traffic lights, road markings, and objects. Roads, lanes, traffic signs, and traffic lights are validated in terms of the point cloud’s relative accuracy. For example, the position of a digitized lane line is compared with the lane line’s position in the point cloud; the difference between the two is measured to determine whether the vector map meets the relative accuracy requirement (Fig. 13).

Fig. 13
figure 13

Schematic of the lane line relative accuracy validation

Full size image

Object and road markings are validated in terms of absolute accuracy during a field survey. For example, the corners of the digitized features of a cross walk are compared with field survey results, and the measured difference determines whether the vector map meets the absolute accuracy requirement. A photograph of a field survey validation is displayed in Fig. 14, and an image depicting absolute accuracy validation is presented in Fig. 15.

Fig. 14
figure 14

Photograph of an absolute accuracy validation field survey

Full size image
Fig. 15
figure 15

Cross walk absolute accuracy validation

Full size image

All of the attributes recorded in an HD maps are validated with point clouds and images obtained from MMS to ensure that these attributes are in accordance with the “Verification and Validation Guidelines for HD Maps” and “HD Maps Data Contents and Formats Standard,” published by the MOI (HD maps data contents and formats standard, 2020; Verification and Validation Guideline for HD Maps, 2020), as shown in Fig. 16.

Fig. 16
figure 16

Schematic of attribute validation

Full size image

2.6 Simulation of Taiwan HD maps with the OpenDRIVE+extension format

OpenDRIVE is an open format designed for autonomous driving simulations. Roads, lanes, lank markings, and other attributes are defined with XML. The XML format is an open and widely accessible format, improving interoperability and facilitating conversion between different formats (OpenDRIVE, 2019). Taiwan HD maps are HD maps in the OpenDRIVE+Extension format based on OpenDRIVE 1.5. Consistent with the structure of OpenDRIVE, the extension includes elements unique to traffic in Taiwan.

Roads and lane elements are the core of Taiwan HD maps. The OpenDRIVE format differs from SHP files in that it encourages a mathematical construction of lane models. Connectivity between different lane sections is a key feature for high-performance autonomous driving; thus, maps in the OpenDRIVE format are more suitable for this application. A simulation was performed using the self-driving simulation software Virtual Test Drive (VTD) developed by MSC Software (the organization maintaining OpenDRIVE). The simulation was performed in three steps: Format validation, 3D visualization, and Self-driving simulation.

2.6.1 Format validation

The OpenDRIVE file was first validated; no error messages should occur when an OpenDRIVE file is imported into VTD software (Fig. 17).

Fig. 17
figure 17

Successful OpenDRIVE file import into VTD

Full size image

2.6.2 3D visualization

The connections between lane sections (Fig. 18) and road markings (Fig. 19) were validated using the 3D visualization.

Fig. 18
figure 18

Lane connection validation

Full size image
Fig. 19
figure 19

Road marking validation

Full size image

2.6.3 Self-driving simulation

A simulation was performed to determine whether self-driving vehicles could navigate various lane sections smoothly without leaving the roadway (Fig. 20).

Fig. 20
figure 20

Potential self-driving simulation outcomes

Full size image

3 Challenges of HD maps production

HD maps is a key technology for autonomous vehicle development. The quality of the produced HD maps affects the development and safety of self-driving technology. Achieving the goal of autonomous vehicles is challenging due to the complex traffic environment. Therefore, a standardized the HD maps specification must be created to facilitate integrating data from various driving scenarios and reduce the costs of development (Liu et al., 2020).

HD maps were produced for several self-driving vehicle testing areas in Taiwan (Fig. 21). All of these maps were validated using the aforementioned procedures. Producing HD maps is challenging because the process is expensive. In addition to the cost of field surveys, producing vector maps requires manual digitization and thus has substantial labor costs. Solutions that can reduce the cost of producing HD maps are urgently needed. Establishing the aforementioned industry standards and specifications may reduce these costs. Moreover, developing automated HD maps production and conversion tools can substantially increase the efficiency of HD maps production. MOI and National Cheng Kung University are participating in a joint project with Nagoya University to develop a semiautomatic production platform for HD maps of Taiwan. These organizations aim to increase the efficiency of the time-consuming manual digitization procedure and develop editing tools for OpenDRIVE files. Maps produced with this semiautomatic production platform are expected to meet the HD maps standards but cost less to produce. Fig. 22 presents a graphic overview of the strategy of reducing the cost of producing HD maps.

Fig. 21
figure 21

HD maps demonstration area in Taiwan

Full size image
Fig. 22
figure 22

Strategy of reducing HD maps production costs

Full size image

4 Conclusion

The era of self-driving vehicles will occur soon, and information and communications technology is developing rapidly. HD maps play an indispensable role in this process. Standardizing the verification and validation procedures of these maps ensures that their quality is sufficient for use by autonomous vehicles. Increasing the efficiency and decreasing the cost of producing HD maps are key to ensuring the availability of high-quality HD maps. Future maps are expected to be produced using data acquired by MMS or autonomous vehicles. The use of automated or semiautomated production technology and the standardized verification and validation process will enable the efficient production of HD maps, completing the HD maps production chain.