High-definition Maps Research Articles

Improving Depth Estimation Using Map-Based Depth Priors

Accurate scene depth is fundamental for robot scene understanding as it adds spatial reasoning. However, accurate scene depth often comes at the cost of expensive additional depth sensors. In this letter, we propose to use map-based depth data as an additional input instead of expensive depth sensors. Such an approach is especially appealing in autonomous driving since map-based depth is commonly available from high-definition maps. To validate this approach, we propose a mapping method that works with common autonomous driving datasets and allows for precise localization using a mix of GNSS-INS and image-based techniques. Furthermore, we propose an entirely learnable three-stage network that handles foreground-background mismatches between the map-based prior depth and the actual scene. Finally, we validate the performance of our method in comparison to several baseline methods and SOTA depth completion methods receiving map-based depth as an input. Our method significantly outperforms these methods both in quantitative and qualitative results. Moreover, our method achieves better metric-scale predictions compared to image-only approaches.

A training dataset for semantic segmentation of urban point cloud map for intelligent vehicles

This paper presents a training dataset of point-wisely labeled global point cloud map (GPCM) in urban area. The labeled GPCM dataset can be used to train machine learning models for point cloud semantic segmentation (PCSS). The trained PCSS machine learning model can classify the semantic class of urban GPCM point-wisely to build a semantic GPCM (SGPCM). The SGPCM can be used to enhance the functionality of many intelligent and autonomous vehicle applications, such as localization, map-based perception, and high-definition (HD) map processing. To maximize the utility of the training dataset for the applications, the semantic classes are designed taking into account the requirements of intelligent and autonomous vehicles in urban area. Therefore, the classes contain urban-related objects, such as road surface markers, curba, and in-tunnel objects, which were not included in other previous datasets. First step of construction process for the PCSS training dataset is a generation of the GPCM in the urban area by using a mobile mapping system (MMS). The points in the generated GPCM with MMS contain information about the position (X, Y, Z) and the surface (R, G, B, I). After building the unlabeled GPCM, each point is labeled with the designed semantic classes thorough semi-automatic and manual labeling. The labeled SGPCM was used to train various state-of-the-art PCSS models to verify the effectiveness of SGPCM train-set for the intelligent and autonomous vehicles.

A graph attention network for road marking classification from mobile LiDAR point clouds

The category of road marking is a crucial element in Mobile laser scanning systems’ (MLSs) applications such as intelligent traffic systems, high-definition maps, location and navigation services. Due to the complexity of road scenes, considerable and various categories, occlusion and uneven intensities in MLS point clouds, finely road marking classification is considered as the challenging work. This paper proposes a graph attention network named GAT_SCNet to simultaneously group the road markings into 11 categories from MLS point clouds. Concretely, the proposed GAT_SCNet model constructs serial computable subgraphs and fulfills a multi-head attention mechanism to encode the geometric, topological, and spatial relationships between the node and neighbors to generate the distinguishable descriptor of road marking. To assess the effectiveness and generalization of the GAT_SCNet model, we conduct extensive experiments on five test datasets of about 100 km in total captured by different MLS systems. Three accuracy evaluation metrics: average Precision, Recall, and F1 of 11 categories on the test datasets exceed 91%, respectively. Accuracy evaluations and comparative studies show that our method has achieved a new state-of-the-art work on road marking classification, especially on similar linear road markings like stop lines, zebra crossings, and dotted lines.

Traffic Landmark Matching Framework for HD-Map Update: Dataset Training Case Study

High-definition (HD) maps determine the location of the vehicle under limited visibility based on the location information of safety signs detected by sensors. If a safety sign disappears or changes, incorrect information may be obtained. Thus, map data must be updated daily to prevent accidents. This study proposes a map update system (MUS) framework that maps objects detected by a road map detection system and the object present in the HD map. Based on traffic safety signs notified by the Korean National Police Agency, 151 types of objects, including traffic signs, traffic lights, and road markings, were annotated manually and semi-automatically. Approximately 3,000,000 annotations were trained based on the you only look once (YOLO) model, suitable for real-time detection by grouping safety signs with similar properties. The object coordinates were then extracted from the mobile mapping system point cloud, and the detection location accuracy was verified by comparing and evaluating the center point of the object detected in the MUS. The performance of the groups with and without specified properties was compared and their effectiveness was verified based on the dataset configuration. A model trained with a Korean road traffic dataset on our testbed achieved a group model of 95% mAP and no group model of 70.9% mAP.

Summarizing large scale 3D mesh for urban navigation

Cameras have become increasingly common in vehicles, smartphones, and advanced driver assistance systems. The areas of application of these cameras in the world of intelligent transportation systems are becoming more and more varied: pedestrian detection, line crossing detection, navigation, …A major area of research currently focuses on mapping that is essential for localization and navigation. However, this step generates an important problem of memory management. Indeed, the memory space required to accommodate the map of a small city is measured in tens gigabytes. In addition, several providers today are competing to produce High-Definition (HD) maps. These maps offer a rich and detailed representation of the environment for highly accurate localization. However, they require a large storage capacity and high transmission and update costs. To overcome these problems, we propose a solution to summarize this type of map by reducing the size while maintaining the relevance of the data for navigation based on vision only. The summary consists in a set of spherical images augmented by depth and semantic information and allowing to keep the same level of visibility in every directions. These spheres are used as landmarks to offer guidance information to a distant agent. They then have to guarantee, at a lower cost, a good level of precision and speed during navigation. Some experiments on real data demonstrate the feasibility for obtaining a summarized map while maintaining a localization with interesting performances.

Object Detection of Road Facilities Using YOLOv3 for High-definition Map Updates

Object Detection of Road Facilities Using YOLOv3 for High-definition Map Updates

Bending the Curve of HD Maps Production for Autonomous Vehicle Applications in Taiwan

Mapping technologies have improved over time, and autonomous driving techniques have advanced substantially over recent decades. High-definition (HD) maps are key for autonomous driving because of their accurate and rich interpretations of road scenes. HD maps provide information about road features, such as lane lines, centerlines, traffic signs, and traffic lights, to help autonomous vehicles navigate safely. HD maps have three major challenges: the standardization of the format of HD maps, conversion between map formats, and lack of techniques for automated HD map generation. These issues influence the costs of HD maps. Therefore, this study proposes strategies to overcome these challenges as well as control the cost with the support of the Ministry of the Interior in Taiwan. We established relevant HD map standards and guidelines to standardize the HD map production procedure. Additionally, we contribute to developing semi-automated HD map production tool to enhance the efficiency of HD map production. Another contribution is to develop HD map format conversion tool to satisfy the map requirement for different end-user. This project not only promotes the development of the Taiwanese autonomous driving industry but also increases its international competitiveness.

Deep Reinforcement Learning-based Integrated Control of Hybrid Electric Vehicles Driven by High Definition Map in Cloud Control System

Deep Reinforcement Learning-based Integrated Control of Hybrid Electric Vehicles Driven by High Definition Map in Cloud Control System

Safe Intersection Management With Cooperative Perception for Mixed Traffic of Human-Driven and Autonomous Vehicles

Autonomous driving systems are highly expected to be used on public roads to improve traffic throughput and road safety, but it will likely take a long transition period before all human-driven vehicles can be replaces with autonomous vehicles. Hence, CAVs have to safely cooperate with the surrounding human drivers, in order to gain the benefits of autonomous driving technologies during such transition periods. In this paper, we present a Distributed Synchronous Intersection Protocol (DSIP) and a Cooperative Perception-based High-Definition Map (CP-HD Map) for mixed traffic environments, in which autonomous vehicles co-exist with human-driven vehicles. First, in DSIP, each CAV utilizes dynamic decision-making mechanisms to adaptively change the vehicle behaviors based on the surrounding environments by using vehicle states and \textit{vehicle mode}. In addition, in CP-HD Map, each CAV uses Vehicle-to-Vehicle (V2V) communications to share the information of detected objects to improve the road safety in the mixed traffic environments. Under these protocols, human-driven vehicles simply follow the traffic lights just like they do today. Finally, we show that DSIP and CP-HD Map increase the traffic throughput around the road intersections when we compared to existing signalized intersections and other V2V communications-based protocols.

A novel method for calibration and verification of roadside millimetre‐wave radar

Roadside perception systems commonly rely on millimetre-wave radars that are stationary. Obtaining global coordinates of radar detection targets requires calibration of radars installed on roadways, which is dangerous if performed via conventional methods. Hazards such as live traffic, particularly on expressways, make safe calibration difficult without road closures. To avoid these issues, this paper proposes a novel calibration method for roadside millimetre-wave radar by using high-definition maps to provide global positioning for radar detection targets, combined with vehicle trajectory data from live traffic. After clustering collected traffic data, a polynomial curve fit provides the shape of lane centrelines. Calibration parameters for the relative-to-global position mapping are then obtained by matching fitted curve sample points to points in high-definition maps using non-linear optimization. The calibration results are verified experimentally using a connected vehicle and roadside unit on an urban street, demonstrating that the proposed method meets lane-level precision requirements for V2X applications. This data-driven method borrows live traffic for calibration and hence is safer and more convenient compared to existing calibration methods.

Study on Relative Accuracy and Verification Method of High-Definition Maps for Autonomous Driving

High-definition maps (HDM) for autonomous driving (AD) are an important component of AD systems. HDMs accurately provide a priori information, including lane lines, and road signs, for AD systems. It is an important task to make a reasonable accuracy assessment of the HDM. The current methods for relative accuracy evaluation of general maps in the field of mapping are not fully applicable to HDMs. In this study, a method based on point set alignment and resampling is used to evaluate the relative accuracy of lane lines, and experiments are conducted based on relevant real HDM data. The results show that the relative accuracy of the lane lines is more detailed and relevant than the traditional method. This has implications for the quality control of HDM production.

A Lane Level Bi-Directional Hybrid Path Planning Method Based on High Definition Map

A global reference path generated by a path search algorithm based on a road-level driving map cannot be directly used to complete the efficient autonomous path-following motion of autonomous vehicles due to the large computational load and insufficient path accuracy. To solve this problem, this paper proposes a lane-level bidirectional hybrid path planning method based on a high-definition map (HD map), which effectively completes the high-precision reference path planning task. First, the global driving environment information is extracted from the HD map, and the lane-level driving map is constructed. Real value mapping from the road network map to the driving cost is realized based on the road network information, road markings, and driving behavior data. Then, a hybrid path search method is carried out for the search space in a bidirectional search mode, where the stopping conditions of the search method are determined by the relaxation region in the two search processes. As the search process continues, the dimension of the relaxation region is updated to dynamically adjust the search scope to maintain the desired search efficiency and search effect. After the completion of the bidirectional search, the search results are evaluated and optimized to obtain the reference path with the optimal traffic cost. Finally, in an HD map based on a real scene, the path search performance of the proposed algorithm is compared with that of the simple bidirectional Dijkstra algorithm and the bidirectional BFS search algorithm. The results show that the proposed path search algorithm not only has a good optimization effect, but also has a high path search efficiency.

Seamless Vehicle Positioning by Lidar-GNSS Integration: Standalone and Multi-Epoch Scenarios

Accurate and seamless vehicle positioning is fundamental for autonomous driving tasks in urban environments, requiring the provision of high-end measuring devices. Light Detection and Ranging (lidar) sensors, together with Global Navigation Satellite Systems (GNSS) receivers, are therefore commonly found onboard modern vehicles. In this paper, we propose an integration of lidar and GNSS code measurements at the observation level via a mixed measurement model. An Extended Kalman-Filter (EKF) is implemented to capture the dynamic of the vehicle movement, and thus, to incorporate the vehicle velocity parameters into the measurement model. The lidar positioning component is realized using point cloud registration through a deep neural network, which is aided by a high definition (HD) map comprising accurately georeferenced scans of the road environments. Experiments conducted in a densely built-up environment show that, by exploiting the abundant measurements of GNSS and high accuracy of lidar, the proposed vehicle positioning approach can maintain centimeter-to meter-level accuracy for the entirety of the driving duration in urban canyons.

FEVA: A Federated Video Analytics Architecture for Networked Smart Cameras

Video analytics using networked smart cameras has become a core function for many applications including surveillance, object detection, AR/VR, and so on. In recent years, a number of architectures have been proposed to organize the computing and networking resources of cloud and edge cameras to collectively complete an analytics task (e.g., 3D reconstruction, multi-view re-identification). Unfortunately, in many applications, image sharing can lead to privacy con-cerns. One example is the high definition map (HD map) for autonomous driving. An HD map has a highly dynamic layer of real-time objects. Vehicles can collectively contribute videos from their onboard cameras to construct such a layer, but the video images can contain private information (e.g., the license plate numbers of front cars). In this article, we propose FEVA, a new federat-ed video analytics architecture. Intrinsically, FEVA keeps the video image data local to the edge for analytics and transmits the analytics results to the cloud for aggregation. FEVA partitions the video analytics computing tasks in a way that is priva-cy-preserving and maximizes the overall analytics accuracy under the computing and communication resource constraints of the edge devices. We show how FEVA can be used in practice by a case study using FEVA to support a video analyt-ics application on multi-view vehicles 3D reconstruction. We implement FEVA by extending the open source platform TensorFlow Federated from Google. We deploy our case in an environment with four Amazon OeepLens cameras. Our eval-uation shows that FEVA can protect privacy while effectively increasing the accuracy of the video analytics application.

Key technologies of multi-agent collaborative high definition map construction

Key technologies of multi-agent collaborative high definition map construction

A generalized data model of high definition maps

A generalized data model of high definition maps

Urban Flood Inundation Simulation Based on High-Precision 3D Modeling

Kim, J.O. and Lee, J.K., 2021. Urban flood inundation simulation based on high-precision 3D modeling. In: Lee, J.L.; Suh, K.-S.; Lee, B.; Shin, S., and Lee, J. (eds.), Crisis and Integrated Management for Coastal and Marine Safety. Journal of Coastal Research, Special Issue No. 114, pp. 454–458. Coconut Creek (Florida), ISSN 0749-0208. Flooding is a frequently occurring climate-related natural disaster that affects several people worldwide. In recent years, severe natural disasters have occurred due to unusual weather, and the damage caused by such disasters has increased significantly because of high population densities and the presence of large buildings in coastal urban areas. The spatial characteristics of urban floods are becoming more complex due to widespread changes in land use, such as in the construction of buildings, roads, and drainage networks. In particular, factors that affect urban flood modeling include the location and geometry of buildings. Hydrological models can adequately simulate both surface and subsurface processes at the watershed level; however, most models cannot accurately simulate urban flooding without 3D city modeling. A 3D simulation of computation results is thus necessary for practical applications. In recent years, flood simulation technology has been used to build 3D virtual environments based on data such as high definition (HD) maps, LiDAR point clouds, and drone images. The primary objective of this study was to develop and simulate an urban flood based on 3D city modeling. The 3D city model was tested for a coastal city in the Republic of Korea. Results of case studies indicate a reasonable match between the observed and modeled flood spatial extents and highlight the importance of considering 3D city models in urban flood simulations.

Topo-Boundary: A Benchmark Dataset on Topological Road-Boundary Detection Using Aerial Images for Autonomous Driving

Road-boundary detection is important for autonomous driving. It can be used to constrain autonomous vehicles running on road areas to ensure driving safety. Compared with online road-boundary detection using on-vehicle cameras/Lidars, offline detection using aerial images could alleviate the severe occlusion issue. Moreover, the offline detection results can be directly employed to annotate high-definition (HD) maps. In recent years, deep-learning technologies have been used in offline detection. But there still lacks a publicly available dataset for this task, which hinders the research progress in this area. So in this letter, we propose a new benchmark dataset, named Topo-boundary, for offline topological road-boundary detection. The dataset contains 25,295 1000×1000-sized 4-channel aerial images. Each image is provided with 8 training labels for different sub-tasks. We also design a new entropy-based metric for connectivity evaluation, which could better handle noises or outliers. We implement and evaluate 3 segmentation-based baselines and 5 graph-based baselines using the dataset. We also propose a new imitation-learning-based baseline which is enhanced from our previous work. The superiority of our enhancement is demonstrated from the comparison. The dataset and our-implemented code for the baselines are available at https://tonyxuqaq.github.io/Topo-boundary/.

Wavelet Transform-Based High-Definition Map Construction From a Panoramic Camera

Wavelet Transform-Based High-Definition Map Construction From a Panoramic Camera

Virtual Target-Based Longitudinal Motion Planning of Autonomous Vehicles at Urban Intersections: Determining Control Inputs of Acceleration With Human Driving Characteristic-Based Constraints

This article describes the development and implementation of virtual target-based longitudinal motion planning of autonomous vehicles at urban intersections ensuring safety and ride comfort. In this study, virtual targets are designed to cope with oncoming vehicles in the blind zone at the intersection for safety. The true field of view (FOV) of cognitive sensors and the virtual target states are constructed based on the sensor specification and intersection road information from a high-definition (HD) map. The future states and intention of sensor-detected targets are inferred and predicted with an interacting multiple model (IMM) filter. The local IMM filters are employed with an intelligent driver model (IDM). Based on predicted target states, two driving modes-"cross" and "stop"-under three different intersection stages-"approach," "intersection in," and "intersection out"-are determined. The model predictive control (MPC) is formulated to determine the control inputs of acceleration with human driving characteristics-based constraints. The proposed algorithm is evaluated through simulation to indicate the effectiveness of the virtual target. The suggested motion planning has been implemented on an autonomous driving vehicle and tested on urban roads.