Kai-Wei Chiang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Kai-Wei Chiang has a total of 138 co-authors. Here are the top co-authors by number of shared publications:
- Meng-Lun Tsai
Department of Geomatics, National Cheng Kung University, Taiwan
11 shared publications - Naser El-Sheimy
Department of Geomatics Engineering, University of Calgary, 2500 University Drive NW, Calgary, AB T2N 1N4, Canada
8 shared publications Meng-Lun Tsai
Department of Geomatics, National Cheng Kung University (NCKU), Tainan, Taiwan
7 shared publications- Surachet Srinara
Department of Civil Engineering, University of Phayao, Phayao 56000, Thailand
6 shared publications - Hsiu-Wen Chang
Department of Geomatics, National Cheng-Kung University, 1 University Road, Tainan 701, Taiwan
5 shared publications - Thanh Trung Duong
Department of Geomatics and Land-administration, Hanoi University of Mining and Geology, Hanoi 122000, Vietnam
5 shared publications
Latest Publications
- Open Access
A Database-Centered Geospatial Infrastructure for Autonomous Vehicle Digital Twins: A Framework Integrating Semantic ODD Layers, Runtime Telemetry, and 3D Urban Context
Autonomous Vehicle (AV) Digital Twins require persistent geospatial infrastructures that connect semantic, spatial, temporal, environmental, road-reference, and 3D urban information across simulation and visualization. This study proposes a database-centered framework integrating PostGIS-managed Operational Design Domain (ODD) layers, runtime telemetry, CityGML-based 3D context, an XODR-derived road reference, live environmental observations, and service-mediated delivery through CARLA, FastAPI/WebSocket, Mission Control, and Cesium. Evaluation comprised 30 Route A runs across six background-vehicle densities (0–40) with five paired-seed repetitions per density, plus five Route B repetitions for cross-route functional repeatability. Route A generated 174,214 source states and achieved 100% database-write success, with mean run-level write latency of 1.950 ± 0.055 ms. Mean REST V2V round-trip time increased from 1.388 ms at BG = 0 to 2.379 ms at BG = 40, while background-actor population maintenance averaged 99.975% across nonzero-density runs. Multi-layer ODD overlap occurred in 10.55% of Route A states, with up to three simultaneous contexts. All evaluated road-reference states remained internally ON TRACK under the implementation-specific diagnostic. Route B reproduced the core persistence, semantic, spatial, and traceability functions under a distinct route-exposure pattern. These results support the framework as a reproducible geospatial backbone for AV-DT simulation and analysis within the tested simulation and hardware conditions.
Open AccessA Database-Centered Geospatial Infrastructure for Autonomous Vehicle Digital Twins: A Framework Integrating Semantic ODD Layers, Runtime Telemetry, and 3D Urban Context
Autonomous Vehicle (AV) Digital Twins require persistent geospatial infrastructures that connect semantic, spatial, temporal, environmental, road-reference, and 3D urban information across simulation and visualization. This study proposes a database-centered framework integrating PostGIS-managed Operational Design Domain (ODD) layers, runtime telemetry, CityGML-based 3D context, an XODR-derived road reference, live environmental observations, and service-mediated delivery through CARLA, FastAPI/WebSocket, Mission Control, and Cesium. Evaluation comprised 30 Route A runs across six background-vehicle densities (0–40) with five paired-seed repetitions per density, plus five Route B repetitions for cross-route functional repeatability. Route A generated 174,214 source states and achieved 100% database-write success, with mean run-level write latency of 1.950 ± 0.055 ms. Mean REST V2V round-trip time increased from 1.388 ms at BG = 0 to 2.379 ms at BG = 40, while background-actor population maintenance averaged 99.975% across nonzero-density runs. Multi-layer ODD overlap occurred in 10.55% of Route A states, with up to three simultaneous contexts. All evaluated road-reference states remained internally ON TRACK under the implementation-specific diagnostic. Route B reproduced the core persistence, semantic, spatial, and traceability functions under a distinct route-exposure pattern. These results support the framework as a reproducible geospatial backbone for AV-DT simulation and analysis within the tested simulation and hardware conditions.
- Open Access
- 28 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Intelligent Transportation Systems
- Vol. 7, 1209-1222
AIO-NAV: INS-Centric All-In-One Navigation Framework Integrating Mono Camera, GNSS, and HD Map for Large-Scale Autonomous Driving
Current Visual-Inertial Odometry (VIO) systems degrade rapidly when visual input is compromised, and fail to fully exploit the observability of industrial-grade Inertial Measurement Units (IMUs). Unlike these visual-centric approaches, this paper presents AIO-NAV, an Inertial Navigation System (INS)-centric All-In-One Navigation framework that addresses these limitations. This framework treats INS as the backbone and integrates aiding sources including Global Navigation Satellite System (GNSS), monocular Visual Odometry (VO), and High-Definition (HD) map. All sensors are fused into a single navigator via a unified Extended Kalman Filter (EKF), which employs a modular design. Key update mechanisms are proposed to enable each aiding source to be independently activated or deactivated while maintaining stable operation. Moreover, each module can be easily maintained or replaced with improved solutions. A proposed invariant VO aiding module substitutes the traditional odometer by providing full 3D relative motion measurements instead of distance-only constraints. HD map-based point cloud registration provides advanced aiding, replacing loop-closure mechanisms for absolute pose corrections when the map is available. Comprehensive experiments in GNSS-challenging environments demonstrate that AIO-NAV significantly outperforms benchmarks, including both system-level (Xsens Vision Navigator) and algorithm-level (VINS-Fusion) comparisons. The system achieves a 3D root mean square error (RMSE) of 1.510 m compared to 5.151 m and 4.965 m, respectively. With HD map aiding, AIO-NAV-HDMAP achieves sub-meter level accuracy (0.795 m maximum error), making it suitable for high-level autonomous driving applications.
- 28 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Intelligent Transportation Systems
- Vol. 7, 1209-1222
Open AccessAIO-NAV: INS-Centric All-In-One Navigation Framework Integrating Mono Camera, GNSS, and HD Map for Large-Scale Autonomous Driving
Current Visual-Inertial Odometry (VIO) systems degrade rapidly when visual input is compromised, and fail to fully exploit the observability of industrial-grade Inertial Measurement Units (IMUs). Unlike these visual-centric approaches, this paper presents AIO-NAV, an Inertial Navigation System (INS)-centric All-In-One Navigation framework that addresses these limitations. This framework treats INS as the backbone and integrates aiding sources including Global Navigation Satellite System (GNSS), monocular Visual Odometry (VO), and High-Definition (HD) map. All sensors are fused into a single navigator via a unified Extended Kalman Filter (EKF), which employs a modular design. Key update mechanisms are proposed to enable each aiding source to be independently activated or deactivated while maintaining stable operation. Moreover, each module can be easily maintained or replaced with improved solutions. A proposed invariant VO aiding module substitutes the traditional odometer by providing full 3D relative motion measurements instead of distance-only constraints. HD map-based point cloud registration provides advanced aiding, replacing loop-closure mechanisms for absolute pose corrections when the map is available. Comprehensive experiments in GNSS-challenging environments demonstrate that AIO-NAV significantly outperforms benchmarks, including both system-level (Xsens Vision Navigator) and algorithm-level (VINS-Fusion) comparisons. The system achieves a 3D root mean square error (RMSE) of 1.510 m compared to 5.151 m and 4.965 m, respectively. With HD map aiding, AIO-NAV-HDMAP achieves sub-meter level accuracy (0.795 m maximum error), making it suitable for high-level autonomous driving applications.
- Open Access
- 15 April 2026
- Published by MDPI AG in Engineering Proceedings
- Vol. 126 (1) , 50
UAV Position Tracking with Ground Cameras
The use of Unmanned Aerial Vehicles (UAVs) has become quite popular in several applications during the last few years. Their spread is motivated by the flexibility of usage of UAVs and by their ability to automatically execute several tasks, mostly thanks to the availability of Global Navigation Satellite Systems (GNSSs), which usually allow reliable outdoor localization of aerial vehicles. However, the extension of task automatic execution indoors, and in other challenging working conditions for the GNSS, requires an alternative positioning system able to compensate for the unreliability or unavailability of GNSS in those cases. To this end, additional sensors are usually considered. Among them, cameras are probably the most popular ones. The most common case of a vision-based positioning system is a camera mounted on a moving platform used to determine its ego-motion in a dead-reckoning approach, i.e., visual odometry. Although this solution is affordable and does not require the installation of any infrastructure, it enables absolute positioning of the camera, i.e., of the UAV, only if certain landmarks, with known position, are visible in the flying area. In contrast, this work considers the use of external cameras installed in the flying area to track the UAV movements. This approach is similar to the one implemented in motion capture systems as well, where a set of static cameras is used to triangulate some target positions using calibrated cameras. Instead, this work investigates the use of vision and machine learning tools to (i) extract the UAV position from each video frame and (ii) estimate its 3D position. Estimation of the 3D UAV position is performed with a single camera, exploiting machine learning tools in order to avoid the need for camera calibration. Performance analysis is provided for a dataset collected at the Agripolis campus of the University of Padua.
- 15 April 2026
- Published by MDPI AG in Engineering Proceedings
- Vol. 126 (1) , 50
Open AccessUAV Position Tracking with Ground Cameras
The use of Unmanned Aerial Vehicles (UAVs) has become quite popular in several applications during the last few years. Their spread is motivated by the flexibility of usage of UAVs and by their ability to automatically execute several tasks, mostly thanks to the availability of Global Navigation Satellite Systems (GNSSs), which usually allow reliable outdoor localization of aerial vehicles. However, the extension of task automatic execution indoors, and in other challenging working conditions for the GNSS, requires an alternative positioning system able to compensate for the unreliability or unavailability of GNSS in those cases. To this end, additional sensors are usually considered. Among them, cameras are probably the most popular ones. The most common case of a vision-based positioning system is a camera mounted on a moving platform used to determine its ego-motion in a dead-reckoning approach, i.e., visual odometry. Although this solution is affordable and does not require the installation of any infrastructure, it enables absolute positioning of the camera, i.e., of the UAV, only if certain landmarks, with known position, are visible in the flying area. In contrast, this work considers the use of external cameras installed in the flying area to track the UAV movements. This approach is similar to the one implemented in motion capture systems as well, where a set of static cameras is used to triangulate some target positions using calibrated cameras. Instead, this work investigates the use of vision and machine learning tools to (i) extract the UAV position from each video frame and (ii) estimate its 3D position. Estimation of the 3D UAV position is performed with a single camera, exploiting machine learning tools in order to avoid the need for camera calibration. Performance analysis is provided for a dataset collected at the Agripolis campus of the University of Padua.
- Open Access
Navigation Error Characteristics of LIO-, VIO-, and RIMU-Assisted INS/GNSS Multi-Sensor Fusion Schemes in a GNSS-Denied Environment
Autonomous vehicles at level 3 and above must maintain high navigation accuracy, particularly in global navigation satellite system (GNSS)-denied environments. The main innovations of this work are threefold. First, we integrate visual inertial odometry (VIO) and light detection and ranging (LiDAR) inertial odometry (LIO) as external updates to mitigate the rapid drift of micro-electromechanical system (MEMS)-based industrial-grade inertial measurement units (IMUs) during long-term GNSS outages. Second, we adopt a redundant IMU (RIMU) approach that fuses multiple low-cost IMUs to reduce sensor noise and improve reliability. Third, we propose a system calibration methodology using both static and dynamic vehicle motion to estimate extrinsic parameters (boresight angles and lever arms) of the sensors, achieving an overall boresight angle root-mean-square error of 0.04 degrees in the simulation. Experiments were conducted under a 7 min GNSS-denied scenario in an underground parking lot, allowing for comparison of the error characteristics of multi-sensor fusion schemes against a navigation-grade reference. The INS/GNSS/LIO framework achieved a two-dimensional root-mean-square position error of 1.22 m (95% position error within 2.5 m), meeting the lane-level (1.5 m) accuracy requirement under a GNSS outage exceeding 7 min without prior maps. In contrast, the RINS/GNSS/VIO framework yielded a 4.71 m 2D mean position error under the same conditions. This paper provides a quantitative comparison of the baseline error characteristics of VIO-, LIO-, and RIMU-assisted INS/GNSS fusion under a GNSS-denied navigation scenario.
Open AccessNavigation Error Characteristics of LIO-, VIO-, and RIMU-Assisted INS/GNSS Multi-Sensor Fusion Schemes in a GNSS-Denied Environment
Autonomous vehicles at level 3 and above must maintain high navigation accuracy, particularly in global navigation satellite system (GNSS)-denied environments. The main innovations of this work are threefold. First, we integrate visual inertial odometry (VIO) and light detection and ranging (LiDAR) inertial odometry (LIO) as external updates to mitigate the rapid drift of micro-electromechanical system (MEMS)-based industrial-grade inertial measurement units (IMUs) during long-term GNSS outages. Second, we adopt a redundant IMU (RIMU) approach that fuses multiple low-cost IMUs to reduce sensor noise and improve reliability. Third, we propose a system calibration methodology using both static and dynamic vehicle motion to estimate extrinsic parameters (boresight angles and lever arms) of the sensors, achieving an overall boresight angle root-mean-square error of 0.04 degrees in the simulation. Experiments were conducted under a 7 min GNSS-denied scenario in an underground parking lot, allowing for comparison of the error characteristics of multi-sensor fusion schemes against a navigation-grade reference. The INS/GNSS/LIO framework achieved a two-dimensional root-mean-square position error of 1.22 m (95% position error within 2.5 m), meeting the lane-level (1.5 m) accuracy requirement under a GNSS outage exceeding 7 min without prior maps. In contrast, the RINS/GNSS/VIO framework yielded a 4.71 m 2D mean position error under the same conditions. This paper provides a quantitative comparison of the baseline error characteristics of VIO-, LIO-, and RIMU-assisted INS/GNSS fusion under a GNSS-denied navigation scenario.
- Open Access
- 29 July 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-G-2, 945-950
Preliminary performance analysis using the PPP-RTK service of the National Land and Mapping Center in Taiwan
Low-cost navigation and positioning systems have become a prominent research field in recent years. In the autonomous vehicle domain, real-time accurate vehicle position determination within lanes is essential for high-level autonomous driving. Regarding real-time performance, while (Precise Point Positioning-Real Time Kinematic)PPP-RTK's corresponding accuracy is not as high as Network-RTK (NRTK), PPP-RTK can provide rapid positioning solutions with a single antenna and single Global Navigation Satellite System(GNSS) receiver. It also features lower bandwidth requirements and unidirectional transmission capabilities than NRTK, allowing more simultaneous users. This makes it a reliable choice when server-side hardware, software, and bandwidth resources are limited. In open-sky environments, autonomous vehicles often rely on GNSS for positioning. However, GNSS signals in urban canyons are susceptible to multi-path effects and NLOS (Non-Line-of-Sight) issues, leading to decreased positioning accuracy. Sensor fusion is the most common solution for maintaining real-time, robust, and high-precision navigation quality across various environments. This research proposes an integration of PPP-RTK and Inertial Navigation System (INS) framework to construct a real-time robust navigation system by integrating PPP-RTK and IMU data using an Extended Kalman Filter (EKF). The experimental equipment includes a tactical-grade Inertial Measurement Unit (IMU) and a low-cost receiver capable of receiving National Land Surveying and Mapping Center (NLSC) correction signals for PPP-RTK functionality. Furthermore, testing was conducted in various challenging outdoor urban environments to validate the proposed algorithm. A high-precision navigation-grade system was used as the reference solution to analyze the accuracy of the proposed real-time navigation system.Additionally, a NovAtel Pwrpak7D-E2 with its internal NovAtel OEM7 receiver utilizing TerraStar service for PPP-RTK implementation was used as a benchmark system to compare the gap between Taiwan's preliminary NLSC PPP-RTK development and existing products. A comparison with NLSC traditional RTK and INS integrated system was also provided.Results show that in GNSS open areas, the 2D Root Mean Square Error (RMSE) between the NLSC PPP-RTK system and EGI-M370 was 2.912 meters lower in accuracy than the benchmark system's 0.585 meters. Although the accuracy of the NLSC PPP-RTK service currently falls below TerraStar's, the system is still in its early development stages. With system parameter adjustment and optimization, it is expected to achieve accuracy comparable to TerraStar, providing a more competitive solution for low-cost autonomous navigation.
- 29 July 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-G-2, 945-950
Open AccessPreliminary performance analysis using the PPP-RTK service of the National Land and Mapping Center in Taiwan
Low-cost navigation and positioning systems have become a prominent research field in recent years. In the autonomous vehicle domain, real-time accurate vehicle position determination within lanes is essential for high-level autonomous driving. Regarding real-time performance, while (Precise Point Positioning-Real Time Kinematic)PPP-RTK's corresponding accuracy is not as high as Network-RTK (NRTK), PPP-RTK can provide rapid positioning solutions with a single antenna and single Global Navigation Satellite System(GNSS) receiver. It also features lower bandwidth requirements and unidirectional transmission capabilities than NRTK, allowing more simultaneous users. This makes it a reliable choice when server-side hardware, software, and bandwidth resources are limited. In open-sky environments, autonomous vehicles often rely on GNSS for positioning. However, GNSS signals in urban canyons are susceptible to multi-path effects and NLOS (Non-Line-of-Sight) issues, leading to decreased positioning accuracy. Sensor fusion is the most common solution for maintaining real-time, robust, and high-precision navigation quality across various environments. This research proposes an integration of PPP-RTK and Inertial Navigation System (INS) framework to construct a real-time robust navigation system by integrating PPP-RTK and IMU data using an Extended Kalman Filter (EKF). The experimental equipment includes a tactical-grade Inertial Measurement Unit (IMU) and a low-cost receiver capable of receiving National Land Surveying and Mapping Center (NLSC) correction signals for PPP-RTK functionality. Furthermore, testing was conducted in various challenging outdoor urban environments to validate the proposed algorithm. A high-precision navigation-grade system was used as the reference solution to analyze the accuracy of the proposed real-time navigation system.Additionally, a NovAtel Pwrpak7D-E2 with its internal NovAtel OEM7 receiver utilizing TerraStar service for PPP-RTK implementation was used as a benchmark system to compare the gap between Taiwan's preliminary NLSC PPP-RTK development and existing products. A comparison with NLSC traditional RTK and INS integrated system was also provided.Results show that in GNSS open areas, the 2D Root Mean Square Error (RMSE) between the NLSC PPP-RTK system and EGI-M370 was 2.912 meters lower in accuracy than the benchmark system's 0.585 meters. Although the accuracy of the NLSC PPP-RTK service currently falls below TerraStar's, the system is still in its early development stages. With system parameter adjustment and optimization, it is expected to achieve accuracy comparable to TerraStar, providing a more competitive solution for low-cost autonomous navigation.
- Open Access
- 29 July 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-G-2, 929-935
Utilizing High-Definition Maps and Simulation Software to Enhance Autonomous Vehicle Safety
The safety of autonomous vehicles (AVs) remains a major challenge, particularly within specific Operational Design Domains (ODDs). While physical testing lacks scalability, computer simulations effectively evaluate AV safety. However, most studies rely on virtual maps generated by simulation software, neglecting real-world complexities and focusing on single intersections rather than broader road networks. To fill the gap, this study lie in the development of a novel simulation workflow combining CARLA and VISSIM simulation softwares using High-definition (HD) maps while reducing data collection efforts and enabling the seamless integration of simulation results. Additionally, this study validates the effectiveness of emergency braking indicators within the AV simulation, confirming their applicability in assessing AV safety performance. The study area conducted near Tainan-City High-Speed Railway Station, Taiwan, analyzes emergency braking events across a network of three intersections and four road segments to identify high-risk zones. HD maps of the study area data were integrated into the CARLA to generate realistic traffic scenarios, while VISSIM modeled traffic flow and signal phases. The results indicate that emergency braking hotspots are concentrated at intersections, turns, and sharp curves near safety islands. This finding suggests that AVs make slower decisions than human drivers due to complex perception and computational models. Additionally, it highlights the importance of HD maps and traffic flow analysis in AV simulations and provides recommendations for improving AV safety, including simplifying road layouts, minimizing sharp turns, and restricting arbitrary lane changes.
- 29 July 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-G-2, 929-935
Open AccessUtilizing High-Definition Maps and Simulation Software to Enhance Autonomous Vehicle Safety
The safety of autonomous vehicles (AVs) remains a major challenge, particularly within specific Operational Design Domains (ODDs). While physical testing lacks scalability, computer simulations effectively evaluate AV safety. However, most studies rely on virtual maps generated by simulation software, neglecting real-world complexities and focusing on single intersections rather than broader road networks. To fill the gap, this study lie in the development of a novel simulation workflow combining CARLA and VISSIM simulation softwares using High-definition (HD) maps while reducing data collection efforts and enabling the seamless integration of simulation results. Additionally, this study validates the effectiveness of emergency braking indicators within the AV simulation, confirming their applicability in assessing AV safety performance. The study area conducted near Tainan-City High-Speed Railway Station, Taiwan, analyzes emergency braking events across a network of three intersections and four road segments to identify high-risk zones. HD maps of the study area data were integrated into the CARLA to generate realistic traffic scenarios, while VISSIM modeled traffic flow and signal phases. The results indicate that emergency braking hotspots are concentrated at intersections, turns, and sharp curves near safety islands. This finding suggests that AVs make slower decisions than human drivers due to complex perception and computational models. Additionally, it highlights the importance of HD maps and traffic flow analysis in AV simulations and provides recommendations for improving AV safety, including simplifying road layouts, minimizing sharp turns, and restricting arbitrary lane changes.
- Open Access
- 28 July 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-G-2, 633-640
Real-Time Precision Navigation Design for Autonomous Vehicle with EGI System
This research presents an innovative system framework, and a multi-sensor integration algorithm aimed at improving the navigation accuracy of autonomous vehicle (AV). This study addresses the shortcomings of the LiDAR-centric approach used in Autoware, a popular open-source platform for self-driving cars. This article presents FalcoNav.AI, a novel INS-centric method, integrating inertial navigation, satellite positioning, Light Detection and Ranging (LiDAR), and High-Definition (HD) maps to enhance navigational performance. An Extended Kalman Filter (EKF) is employed for efficient data fusion and processing of Inertial Navigation System (INS). The performance of system is evaluated in various environments, including open sky, Global Navigation Satellite System (GNSS) challenge and GNSS denied areas, showing significant improvements in navigation accuracy and reliability. The key components of the system include affordable Velodyne VLP-16 and a custom-built Embedded GNSS/INS (EGI) Inertial Navigation System (EGI-370). Experiments demonstrate that the system achieves "Where-in-Lane" level accuracy, highlighting its potential for wide application in autonomous vehicle. This innovation represents a significant advance towards more dependable and precise navigation in a wide range of driving conditions.
- 28 July 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-G-2, 633-640
Open AccessReal-Time Precision Navigation Design for Autonomous Vehicle with EGI System
This research presents an innovative system framework, and a multi-sensor integration algorithm aimed at improving the navigation accuracy of autonomous vehicle (AV). This study addresses the shortcomings of the LiDAR-centric approach used in Autoware, a popular open-source platform for self-driving cars. This article presents FalcoNav.AI, a novel INS-centric method, integrating inertial navigation, satellite positioning, Light Detection and Ranging (LiDAR), and High-Definition (HD) maps to enhance navigational performance. An Extended Kalman Filter (EKF) is employed for efficient data fusion and processing of Inertial Navigation System (INS). The performance of system is evaluated in various environments, including open sky, Global Navigation Satellite System (GNSS) challenge and GNSS denied areas, showing significant improvements in navigation accuracy and reliability. The key components of the system include affordable Velodyne VLP-16 and a custom-built Embedded GNSS/INS (EGI) Inertial Navigation System (EGI-370). Experiments demonstrate that the system achieves "Where-in-Lane" level accuracy, highlighting its potential for wide application in autonomous vehicle. This innovation represents a significant advance towards more dependable and precise navigation in a wide range of driving conditions.
- Open Access
- 28 July 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-G-2, 625-631
The development and application of Taiwan HD Maps for smart mobile mapping technology
The rapid advancement of Intelligent Transport Systems (ITS) has significantly propelled the development of autonomous vehicle technologies in recent years. HD Maps are crucial for autonomous driving technology due to their high accuracy, rich road scene semantic data, real-time traffic conditions, and driving experience information. However, how to produce HD Maps efficiently and meet the accuracy for autonomous vehicles are important challenge. This study proposes related methods to meet the requirement. First one is to publish HD Maps guidelines and standards which are published since 2018, with updated versions scheduled for publication in 2024 and 2025. These documents outline the operating procedures, content format, verification details, and other information to ensure that the final HD Maps maintain consistent accuracy and quality. Second, a semi-automated HD Maps production tool to generate lane lines, stop lines, crosswalks, and other features essential for autonomous vehicle control systems is proposed. This study also defines the procedures and variant detection tools for certified third-party mapping technologies. These tools are designed to improve the efficiency of HD Maps production while reducing costs. Finally, this study creates seamless outdoor and indoor HD Maps in Taiwan which includes closed fields, real roads, and indoor parking lots, supports various application scenarios for autonomous vehicles. This integrated approach not only promotes the advancement of autonomous driving technologies but also drives improvements in mapping methodologies. Establishing a robust HD Maps infrastructure is expected to accelerate the development of fully autonomous vehicles, bringing their widespread deployment closer to reality.
- 28 July 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-G-2, 625-631
Open AccessThe development and application of Taiwan HD Maps for smart mobile mapping technology
The rapid advancement of Intelligent Transport Systems (ITS) has significantly propelled the development of autonomous vehicle technologies in recent years. HD Maps are crucial for autonomous driving technology due to their high accuracy, rich road scene semantic data, real-time traffic conditions, and driving experience information. However, how to produce HD Maps efficiently and meet the accuracy for autonomous vehicles are important challenge. This study proposes related methods to meet the requirement. First one is to publish HD Maps guidelines and standards which are published since 2018, with updated versions scheduled for publication in 2024 and 2025. These documents outline the operating procedures, content format, verification details, and other information to ensure that the final HD Maps maintain consistent accuracy and quality. Second, a semi-automated HD Maps production tool to generate lane lines, stop lines, crosswalks, and other features essential for autonomous vehicle control systems is proposed. This study also defines the procedures and variant detection tools for certified third-party mapping technologies. These tools are designed to improve the efficiency of HD Maps production while reducing costs. Finally, this study creates seamless outdoor and indoor HD Maps in Taiwan which includes closed fields, real roads, and indoor parking lots, supports various application scenarios for autonomous vehicles. This integrated approach not only promotes the advancement of autonomous driving technologies but also drives improvements in mapping methodologies. Establishing a robust HD Maps infrastructure is expected to accelerate the development of fully autonomous vehicles, bringing their widespread deployment closer to reality.
- Open Access
- 10 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Intelligent Transportation Systems
- Vol. 6, 995-1008
AI-Driven Mapping System for Smart Parking Management Applications Using an INS-GNSS-Solid-State LiDAR-Monocular Camera Fusion Engine Empowered by HD Maps
Efficient parking management is crucial in crowded Asian cities to optimize limited road space and parking facilities. The increasing vehicle ownership rate in Taiwan has intensified the demand for street parking, leading to excessive driving in search of available spots and contributing up to 30% of traffic congestion. This paper proposes a low-cost, infrastructure-free outdoor roadside parking management system based on high-definition (HD) map updating. The system fuses data from a solidstate LiDAR (SSL) system, a monocular camera, an inertial navigation system, a GPS, and HD maps followed by deep-learningbased efficient region extraction. The goal was to achieve high accuracy with minimal computational resources and infrastructure costs. The proposed systems performance for dynamic HD map object updating was evaluated through parking management tests. The systems costs were low due to the selection of SSL and monocular cameras. Traditional and novel extrinsic calibration methods were compared in various experiments, and a hardware architecture for precise sensor time synchronization was designed. Software algorithms for accurate image–point-cloud projection were developed to update HD map parking layers. By using normal distribution transform matching of the SSL and HD point cloud maps, navigation performance was achieved to 0.4-meter accuracy. When applied to license plate localization in two experimental scenarios, the mean performance error was approximately 0.48 and 0.62 m.
- 10 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Intelligent Transportation Systems
- Vol. 6, 995-1008
Open AccessAI-Driven Mapping System for Smart Parking Management Applications Using an INS-GNSS-Solid-State LiDAR-Monocular Camera Fusion Engine Empowered by HD Maps
Efficient parking management is crucial in crowded Asian cities to optimize limited road space and parking facilities. The increasing vehicle ownership rate in Taiwan has intensified the demand for street parking, leading to excessive driving in search of available spots and contributing up to 30% of traffic congestion. This paper proposes a low-cost, infrastructure-free outdoor roadside parking management system based on high-definition (HD) map updating. The system fuses data from a solidstate LiDAR (SSL) system, a monocular camera, an inertial navigation system, a GPS, and HD maps followed by deep-learningbased efficient region extraction. The goal was to achieve high accuracy with minimal computational resources and infrastructure costs. The proposed systems performance for dynamic HD map object updating was evaluated through parking management tests. The systems costs were low due to the selection of SSL and monocular cameras. Traditional and novel extrinsic calibration methods were compared in various experiments, and a hardware architecture for precise sensor time synchronization was designed. Software algorithms for accurate image–point-cloud projection were developed to update HD map parking layers. By using normal distribution transform matching of the SSL and HD point cloud maps, navigation performance was achieved to 0.4-meter accuracy. When applied to license plate localization in two experimental scenarios, the mean performance error was approximately 0.48 and 0.62 m.
- 29 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (23) , 37582-37598
Creation and Verification of High-Definition Point Cloud Maps for Autonomous Vehicle Navigation
High-definition (HD) maps have recently become a key piece of technology in autonomous driving. Over the past few years, various methods and sensors, such as those based on inertial navigation system (INS), global navigation satellite system (GNSS), cameras, and light detection and ranging (LiDAR), have been used to develop HD maps. In this study, we developed novel techniques for enhancing the creation and verification of HD point cloud maps. Firstly, a tightly coupled (TC) INS/GNSS-assisted 3D Normal Distribution Transform (NDT)-LiDAR mapping system has been developed. Utilizing an integrated INS/GNSS, the system provides a reliable initial pose, thereby mitigating the issue of divergence in NDT scan matching, particularly when the vehicle operates at high speeds in challenging LiDAR environments. This approach enhances both navigation accuracy and the precision of the point cloud map. Secondly, alternative ground control points (GCPs) have been established as substitutes for conventional techniques, addressing freeway regulations and managing safety concerns. Thirdly, to ensure the desired accuracy for “where-in-lane” positioning in autonomous vehicle applications, the created point cloud map was validated against the criteria outlined by standardized procedures. Overall, our preliminary results indicate that our HD point cloud map meets the positioning accuracy criteria outlined by the Taiwan Association of Information and Communication Standards. Our point density results also indicate that our generated point cloud map can achieve a high degree of accuracy in in-lane positioning for autonomous vehicle navigation.
- 29 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 11 (23) , 37582-37598
Creation and Verification of High-Definition Point Cloud Maps for Autonomous Vehicle Navigation
High-definition (HD) maps have recently become a key piece of technology in autonomous driving. Over the past few years, various methods and sensors, such as those based on inertial navigation system (INS), global navigation satellite system (GNSS), cameras, and light detection and ranging (LiDAR), have been used to develop HD maps. In this study, we developed novel techniques for enhancing the creation and verification of HD point cloud maps. Firstly, a tightly coupled (TC) INS/GNSS-assisted 3D Normal Distribution Transform (NDT)-LiDAR mapping system has been developed. Utilizing an integrated INS/GNSS, the system provides a reliable initial pose, thereby mitigating the issue of divergence in NDT scan matching, particularly when the vehicle operates at high speeds in challenging LiDAR environments. This approach enhances both navigation accuracy and the precision of the point cloud map. Secondly, alternative ground control points (GCPs) have been established as substitutes for conventional techniques, addressing freeway regulations and managing safety concerns. Thirdly, to ensure the desired accuracy for “where-in-lane” positioning in autonomous vehicle applications, the created point cloud map was validated against the criteria outlined by standardized procedures. Overall, our preliminary results indicate that our HD point cloud map meets the positioning accuracy criteria outlined by the Taiwan Association of Information and Communication Standards. Our point density results also indicate that our generated point cloud map can achieve a high degree of accuracy in in-lane positioning for autonomous vehicle navigation.