High-definition Maps Research Articles

LiDAR mini-matching positioning method based on constraint of lightweight point cloud feature map

In the global navigation satellite positioning denial environment, the matching localization based on a high-definition map is a well-recognized method to assist autonomous vehicles in achieving continuous, stable, and reliable navigation and positioning in complex scenes. However, due to the complexity, density, and richness of the typical high-definition point cloud map, there are high technical thresholds in the process of localization production, application, and maintenance. As a result, referring to the idea of traverse survey in geodesy, we design a priori lightweight point cloud feature map structure based on road mileage nodes as the main line to increase the flexibility of map segmentation, splicing, retrieval, and communication transmission. At the same time, an efficient and practical mini-matching positioning method is proposed. Finally, using the real datasets of various urban canyon scenes for verification, the proposed method can achieve stable sub-meter level localization.

High‐definition map automatic annotation system based on active learning

AbstractAs autonomous vehicle technology advances, high‐definition (HD) maps have become essential for ensuring safety and navigation accuracy. However, creating HD maps with accurate annotations demands substantial human effort, leading to a time‐consuming and costly process. Although artificial intelligence (AI) and computer vision (CV) algorithms have been developed for prelabeling HD maps, a significant gap remains in accuracy and robustness between AI‐based methods and traditional manual pipelines. Additionally, building large‐scale annotated datasets and advanced machine learning algorithms for AI‐based HD map labeling systems can be resource‐intensive. In this paper, we present and summarize the Tencent HD Map AI (THMA) system, an innovative end‐to‐end, AI‐based, active learning HD map labeling system designed to produce HD map labels for hundreds of thousands of kilometers while employing active learning to enhance product iteration. Utilizing a combination of supervised, self‐supervised, and weakly supervised learning, THMA is trained directly on massive HD map datasets to achieve the high accuracy and efficiency required by downstream users. Deployed by the Tencent Map team, THMA serves over 1000 labeling workers and generates more than 30,000 km of HD map data per day at its peak. With over 90% of Tencent Map's HD map data labeled automatically by THMA, the system accelerates traditional HD map labeling processes by more than tenfold, significantly reducing manual annotation burdens and paving the way for more efficient HD map production.

A Robust Localization Approach Based on Point Cloud Descriptor and SLAM in Urban Environment

The importance of autonomous driving localization to autonomous driving systems cannot be overstated, reliable and high-precision localization is the focus of research in autonomous driving. Many studies at this stage use multiple sensors to generate high-definition maps for matching and localization. The labor and economic costs required to maintain such maps are often high. This paper proposes a fusion localization framework for simultaneous localization and mapping (SLAM) and matching localization. It could afford high-precision localization results of autonomous vehicles using only light detection and ranging (LiDAR) sensors. Concretely, a novel global point cloud descriptor, named binary scan context (BSC) is proposed, for matching localization. It encodes the structural features in the z-height direction in the point cloud space in a binary way. An efficient two-stage matching strategy is proposed to improve the efficiency of matching. Experiments on public datasets and vehicles have demonstrated the validation and precision of the method.

PRS-HDMC: An online participant recruitment scheme for vehicular crowdsensing-enabled HD map collection

The emergence of vehicular crowdsensing (VCS) brings new opportunities and possibilities for the collection of High-Definition (HD) maps. Unlike traditional mapping methods, VCS leverages ubiquitous vehicles and on-board cameras to collect data inexpensively and efficiently. Though promising, such a data collection mode still faces an unsolved problem of ensuring the availability and integrity of collected map data. To address this issue, this paper proposes PRS-HDMC, a participant recruitment scheme for VCS-enabled HD map collection. Specifically, we first analyze the unique requirements that distinguish HD map acquisition from other sensing tasks. On this basis, a metric participant contribution is proposed to describe the accuracy and content richness of the collected photo. Combining this metric with participant reliability and task coverage, we establish a novel quality of service (QoS) quantification system and develop the participant recruitment optimization model. Then, an improved greedy algorithm is proposed to recruit the appropriate vehicular set online according to their real-time locations, thereby maximizing the system QoS of each task round. Finally, we conduct extensive simulations based on a synthetic dataset under different settings. Experimental results demonstrate that the proposed scheme can guarantee the availability and integrity of collected map data, and outperforms all benchmarks.

Semantic interpretation of raw survey vehicle sensory data for lane-level HD map generation

High-definition (HD) maps provide a complementary source of information for Advanced Driver Assistance Systems (ADAS), allowing them to better understand the vehicle’s surroundings and make more informed decisions. HD maps are also largely employed in virtual testing phases to evaluate the behavior of ADAS components under simulated conditions. With the advent of autonomous sensorized vehicles, raw machine-oriented data will be increasingly available. The proposed pipeline aims to provide a high-level semantic interpretation of raw vehicle sensory data to derive, in an automated fashion, lane-oriented HD maps of the environment. We first present RoadStarNet, a deep learning architecture designed to extract and classify road line markings from imagery data. We show how to obtain a semantic Bird’s-Eye View (BEV) mapping of the extracted road line markings by exploiting frame-by-frame localization information. Then, we present how to progress to a graph-based representation that allows modeling complex road line markings’ structures practically, as this representation can be leveraged to produce a Lanelet2 format HD map. Lastly, we experimentally evaluate the proposed approach in real-world scenarios in terms of accuracy and coverage performance.

Evaluation of Rooftop Photovoltaic Power Generation Potential Based on Deep Learning and High-Definition Map Image

Photovoltaic (PV) power generation is booming in rural areas, not only to meet the energy needs of local farmers but also to provide additional power to urban areas. Existing methods for estimating the spatial distribution of PV power generation potential either have low accuracy and rely on manual experience or are too costly to be applied in rural areas. In this paper, we discuss three aspects, namely, geographic potential, physical potential, and technical potential, and propose a large-scale and efficient PV potential estimation system applicable to rural rooftops in China. Combined with high-definition map images, we proposed an improved SegNeXt deep learning network to extract roof images. Using the national standard Design Code for Photovoltaic Power Plants (GB50797-2012) and the Bass model, computational results were derived. The average pixel accuracy of the improved SegNeXt was about 96%, which well solved the original problems of insufficient finely extracted edges, poor adhesion, and poor generalization ability and can cope with different types of buildings. Leizhou City has a geographic potential of 1500 kWh/m2, a physical potential of 25,186,181.7 m2, and a technological potential of 442.4 MW. For this paper, we innovatively used the Bass Demand Diffusion Model to estimate the installed capacity over the next 35 years and combined the Commodity Diffusion Model with the installed capacity, which achieved a good result and conformed to the dual-carbon “3060” plan for the future of China.

Enhancing vehicle localization by matching HD map with road marking detection

Accurate and robust vehicle localization, a fundamental task for autonomous driving, is still a challenging problem especially in GPS-denied scenarios. Recently, high definition (HD) map suggests a promising solution. However, the matching between online sensed data and HD map is difficult and time-consuming. In our work, road markings are selected as landmarks due to salient appearance features. Based on the detection results of road markings from the vehicle-borne images, the points of edge lines are employed to fit straight lines with RANSAC for outlier removal. The distances between the ego-vehicle and the fitted edge lines can be computed with camera-vehicle calibration in advance. Subsequently, the point-to-line distances from the sensed data are mapped into global linear constraints on the vehicle’s positions with the support of a lane-level HD map, which provides centimeter-level coordinates of road markings. Finally, the distances and the localization from the integrated navigation system (INS) are fused with the proposed linear Kalman filter based on the second-order Markov model (KF-MM2). The proposed method has been verified in different daily driving scenarios. Experimental results demonstrate that our method can achieve good performance with an average localization error of 0.53 m and a standard deviation of 0.17 m.

Lane-Level Map Generation and Management Framework Using Connected Car Data

This study proposes a lane-level map generation and management framework using connected sensor data to reduce the manpower and time required for producing and updating high-definition (HD) maps. Unlike previous studies that relied on the onboard processing capabilities of vehicles to collect map-constructing elements, this study offloads computing for map generation to the cloud, assigning vehicles solely the role of transmitting sensor data. For efficient data collection, we divide the space into a grid format to define it as a partial map and establish the state of each map and its transition conditions. Lastly, tailored to the characteristics of the road elements composing the map, we propose an automated map generation technique and method for selectively collecting data. The map generation method was tested using data collected from actual vehicles. By transmitting images with an average size of 350 KB, implementation was feasible even with the current 5G upload bandwidth. By utilizing 12,545 elements, we were able to achieve a position accuracy and regression RMSE of less than 0.25 m, obtaining 651 map elements to construct the map. We anticipate that this study will help reduce the manpower and time needed for deploying and updating HD maps.

Recent Developments on Drivable Area Estimation: A Survey and a Functional Analysis.

Most advanced autonomous driving systems (ADS) today rely on the prior creation of high-definition maps (HD maps). This process is expensive and needs to be performed frequently to keep up with the changing conditions of the road environment. Creating accurate navigation maps online is an alternative to reduce the cost and broaden the current operational design domains (ODD) of modern ADS. This paper offers a snapshot of the state of the art in drivable area estimation, which is an essential technology to deploy ADS in ODDs where HD maps are limited or unavailable. The proposed review introduces a novel architecture breakdown that fits learning-based and non-learning-based techniques and allows the analysis of a set of impactful and recent drivable area algorithms. In addition to that, complimentary information for practitioners is provided: (i) an assessment of the influence of modern sensing technologies on the task under study and (ii) a selection of relevant datasets for evaluation and benchmarking purposes.

Milestones in Autonomous Driving and Intelligent Vehicles—Part I: Control, Computing System Design, Communication, HD Map, Testing, and Human Behaviors

Interest in autonomous driving (AD) and intelligent vehicles (IVs) is growing at a rapid pace due to the convenience, safety, and economic benefits. Although a number of surveys have reviewed research achievements in this field, they are still limited in specific tasks and lack systematic summaries and research directions in the future. Our work is divided into 3 independent articles and the first part is a Survey of Surveys (SoS) for total technologies of AD and IVs that involves the history, summarizes the milestones, and provides the perspectives, ethics, and future research directions. This is the second part (Part I for this technical survey) to review the development of control, computing system design, communication, High Definition map (HD map), testing, and human behaviors in IVs. In addition, the third part (Part II for this technical survey) is to review the perception and planning sections. The objective of this paper is to involve all the sections of AD, summarize the latest technical milestones, and guide abecedarians to quickly understand the development of AD and IVs. Combining the SoS and Part II, we anticipate that this work will bring novel and diverse insights to researchers and abecedarians, and serve as a bridge between past and future.

Registration Method of Single-Line Point Clouds and Images Based on Depth Learning

In high-definition map production, the point cloud and image data collected by the collection vehicle play a supportive role in high-definition map production. Point cloud data can provide accurate 3D spatial positioning, while image data can provide rich semantic information. The generation of high-definition maps depends on the fusion of multi-sensor data, and sensor registration is a requirement for multi-sensor fusion. The relative position between the LiDAR and the camera is largely resolved by the registration of single-line point clouds with images. Conventional manual registration techniques require a lot of labor and specialized gear for calibration, and they must be recalibrated if the distances between the sensors vary. This work suggests an online, unsupervised, and automatic deep learning-based registration approach for single-line point clouds and images. The technique was evaluated on an internal dataset, and it decreased the average displacement error to 33.22 cm and the average rotation error to 0.54° when the highest displacement error was 1.5 m and the maximum rotation error was 20°.

Conditional Generative Models for Dynamic Trajectory Generation and Urban Driving.

This work explores methodologies for dynamic trajectory generation for urban driving environments by utilizing coarse global plan representations. In contrast to state-of-the-art architectures for autonomous driving that often leverage lane-level high-definition (HD) maps, we focus on minimizing required map priors that are needed to navigate in dynamic environments that may change over time. To incorporate high-level instructions (i.e., turn right vs. turn left at intersections), we compare various representations provided by lightweight and open-source OpenStreetMaps (OSM) and formulate a conditional generative model strategy to explicitly capture the multimodal characteristics of urban driving. To evaluate the performance of the models introduced, a data collection phase is performed using multiple full-scale vehicles with ground truth labels. Our results show potential use cases in dynamic urban driving scenarios with real-time constraints. The dataset is released publicly as part of this work in combination with code and benchmarks.

Probabilistic Semantic Mapping for Autonomous Driving in Urban Environments.

Statistical learning techniques and increased computational power have facilitated the development of self-driving car technology. However, a limiting factor has been the high expense of scaling and maintaining high-definition (HD) maps. These maps are a crucial backbone for many approaches to self-driving technology. In response to this challenge, we present an approach that fuses pre-built point cloud map data with images to automatically and accurately identify static landmarks such as roads, sidewalks, and crosswalks. Our pipeline utilizes semantic segmentation of 2D images, associates semantic labels with points in point cloud maps to pinpoint locations in the physical world, and employs a confusion matrix formulation to generate a probabilistic bird's-eye view semantic map from semantic point clouds. The approach has been tested in an urban area with different segmentation networks to generate a semantic map with road features. The resulting map provides a rich context of the environment that is valuable for downstream tasks such as trajectory generation and intent prediction. Moreover, it has the potential to be extended to the automatic generation of HD maps for semantic features. The entire software pipeline is implemented in the robot operating system (ROS), a widely used robotics framework, and made available.

Automatically Refining Degraded and Occluded Lane Marking Information to Enhance the quality of High Definition Maps for Autonomous Vehicles

Autonomous vehicle technology has been advancing over the past decade with many challenges and obstacles. One obstacle to achieving fully autonomous driving is the feasibility of efficiently producing robust high definition (HD) maps of existing road infrastructure in a sustainable manner. HD maps are digital twins of road infrastructure that include 3-D representation of road features critical to a vehicle’s ability to navigate a road and maintain a consistent trajectory. This includes lane markings, barriers, and road edges. HD maps are often produced by first surveying roads using remote sensing technology. The collected data are then segmented using computer vision and machine learning algorithms to produce an HD map of the features of concern. One common challenge that is specific to extracting lane marking information is that the markings are occasionally occluded or degraded by high traffic volumes. This results in failure to detect lane markings and the existence of significant gaps in the extracted information. These gaps are manually filled in by quality control staff in a tedious and time-consuming process. To help overcome such challenges, this paper proposes a novel algorithm through which Kalman filtering is combined with Bézier curve fitting to automatically refine lane marking information. The proposed algorithm is tested on multiple roads where mobile lidar data were collected and segmentation was first applied using deep learning technology. The proposed algorithm was then used to refine the lane marking information which helped close all the gaps and recreate missing portions of the extracted lane markings.

Lane Information Extraction for High Definition Maps Using Crowdsourced Data

Lane information plays an important role in high-definition (HD) maps because it provides invaluable road information for autonomous vehicles. The most widely used lane extraction method for HD maps is based on mobile mapping systems, and it is prohibitively costly and time-consuming. In this study, we propose a novel approach for lane information extraction based on crowdsourcing vehicles equipped with monocular cameras and global navigation satellite system devices for recording road images and position data. First, we propose a lane mask propagation network to detect the lane markings in images, which are then projected from a perspective space into a three-dimensional space in accordance with the position data. Second, we propose a data management method to store lane information in the cloud data center. An improved density-based spatial clustering of applications with noise clustering algorithm and a gradual fitting algorithm are used to remove the outliers and improve the lane data accuracy. The proposed method is quantitatively evaluated against a real-world HD map produced by a mobile mapping vehicle. The experimental results show that more than 80% of the extracted lane markings meet the accuracy requirements of HD maps. In conclusion, our method can be used as a low-cost and efficient approach for updating the lane information in HD maps for autonomous vehicles.

Trajectory-Based Method for Dividing Lanes of Vehicle Trajectories Collected by Roadside LiDAR

The detailed vehicle trajectories collected by various advanced sensors (e.g., LiDARs, radars, digital cameras) form an essential foundation for in-depth analysis of connected and automated transportation systems, in which lane division of vehicle trajectories is an essential step. Rather than high-definition maps, trajectory-based methods have attracted increasing attention for their merits of low cost and adaptability to map changes in a timely manner. However, they require a priori geometric information about roads which is usually unavailable in practice. To address this problem, we propose a method solely based on trajectory data to derive geometric road features for dividing lanes of trajectories collected in connected environments. The proposed method consists of two stages: (i) to extract the lane lines of the road, the direction of the road centerline is calculated as the weighted arithmetic average of trajectory directions, and an iterative process can keep updating the feature points of road centerline and lane lines based on kernel density estimation of trajectories; (ii) trajectories are divided into lanes by constructing a polygon based on the extracted lane lines. We conducted a field experiment to collect high-precision vehicle trajectories using roadside LiDAR devices and created a high-definition digital map at an urban road without signalized intersection in the city of Chengdu. The experiment based on the empirical data demonstrates the proposed method can accurately divide lanes for 97.83% of trajectory points. In addition, a series of sensitivity analyses show that the proposed method is robust to the noise in trajectory data and the error of calibrating the roadside LiDAR devices.

On the role of lidar measurements in speeding up precise point positioning convergence

Global navigation satellite system (GNSS) and light detection and ranging (lidar) are well known to be complementary for vehicle positioning in urban canyons, where GNSS observations are prone to signal blockage and multi-path. As one of the most common carrier-phase-based precise positioning techniques, precise point positioning (PPP) enables single-receiver positioning as it utilizes state-space representation corrections for satellite orbits and clocks and does not require a nearby reference station. Yet PPP suffers from a long positioning convergence time. In this contribution, we propose to reduce the PPP convergence using an observation-level integration of GNSS and lidar. Lidar measurements, in the form of 3D keypoints, are generated by registering online scans to a pre-built high-definition map through deep learning and are then combined with dual-frequency PPP (DF-PPP) observations in an extended Kalman filter implementing the constant-velocity model that captures the vehicle dynamics. We realize real-time PPP (RT-PPP) in this integration using the IGS real-time service products for vehicle positioning. Comprehensive analyses are provided to evaluate different combinations of measurements and PPP corrections in both static and simulated kinematic modes using data captured by multiple receivers. Experimental results show that the integration achieves cm-level accuracy and instantaneous convergence by using redundant measurements. Accordingly, for classical PPP accuracy of 10 cm and convergence within minutes, respectively, lidar input is only required once every 10 s.

Vehicle to Infrastructure-Based LiDAR Localization Method for Autonomous Vehicles

The localization of autonomous vehicles using light detection and ranging (LiDAR) sensors relies on high-definition (HD) maps, which are essential for accurate positioning. However, the large storage capacity required for HD maps poses challenges for real-time performance. To address this issue, we propose a vehicle to infrastructure (V2I)-based LiDAR localization method. In this approach, real-time HD maps are transmitted to vehicles in the vicinity of the infrastructure, enabling localization without the need for map data. We conducted tests to determine the optimal size of the HD maps and the distance between vehicles and the infrastructure, considering the impact on transmission speed. Additionally, we compared the matching performance between the complete HD map and sub maps received from the infrastructure, to evaluate the effectiveness of our method in a qualitative manner.

Open HD map service model: an interoperable high-Definition map data model for autonomous driving

ABSTRACT With the growth in the vehicle industry, autonomous driving has become a hot topic worldwide and has attracted increasing attention from both industrial and academic sectors. Maps, as pivotal geospatial information carriers, play a vital role in route planning and navigation service. Compared with conventional maps, high-definition (HD) maps possesses higher precision, richer information, and various services and are regarded as critical infrastructure for autonomous driving. However, heterogeneous HD map data standards and models have different characteristics and advantages, and thus they rarely meet all autonomous driving requirements for different driving objectives. This research presents an interoperable map data model, the Open HD Map Service Model (OHDMSM), to provide a reference for HD map development. The designed OHDMSM, which contains three data layers and a set of corresponding interfaces, demonstrates high interoperability for HD map data fusion and application. As a proof of concept, an HD map data system is implemented with all functions following the designed data model and interfaces of OHDMSM. The design and development of OHDMSM data structures, interfaces and systems will benefit data requesting, updating, and interoperation for HD map data worldwide, which can be helpful for developing autonomous driving and intelligent transportation in the Digital Earth.

Trajectory Prediction for Automated Vehicles on Roads With Lanes Partially Covered by Ice or Snow

Snow and ice films obscure lane lines during the winter, leading to inaccurate trajectory predictions and reducing a driver's ability to guide the vehicle. The full use of observable information to continuously infer trajectories is an urgent issue in trajectory prediction tasks. In this paper, a neighboring vehicle trajectory prediction method based on graph neural networks (GNNs) is proposed, the lane line positions and attribute information in high-definition (HD) maps are extracted, and an ice and snow mask mechanism that simulates instances in which lane lines are covered is introduced. Because graph aggregation with a GNN enables information exchange between nodes, a graph query mechanism that integrates local graph information and guides prediction results based on the local states of the nearest observable nodes is proposed. In addition, a multihead attention mechanism projects neighboring states to the graph, describing interactions between vehicles and the state of the traffic flow, and a two-layer graph attention network (GAT) enables information aggregation and describes node correlations. Experiments on the nuScenes dataset show that our prediction method outperforms state-of-the-art prediction systems, including Traj++, SG-NET, and P2T, when lane lines are occluded. Vehicle trajectories on roads covered by ice and snow can be accurately predicted based on observable information.