Construction Methods for an Autonomous Driving Map in an Intelligent Network Environment

By Zhijun Chen

This book provides an overview of constructing advanced Autonomous Driving Maps. It includes coverage of such methods as: fusion target perception (based on vehicle vision and millimeter wave radar), cross-field of view object perception, vehicle motion recognition (based on vehicle road fusion information), vehicle trajectory prediction (based on improved hybrid neural network) and the driving map construction method driven by road perception fusion.

An Autonomous Driving Map is used for optimization of not only for a single vehicle, but also for the entire traffic system.

• Language: English
• Publisher: Elsevier Science
• Release date: Apr 4, 2024
• ISBN: 9780443273179

Zhijun Chen

Dr Chen is the Deputy Director of the Institute of Traffic Information and Intelligent Systems, Intelligent Transportation Systems Research Center, Wuhan University of Technology. His expertise areas include artificial intelligence, image processing, big data mining, vehicle-road collaboration and connected automated driving, intelligent driving, autonomous driving

Book preview

Preface

Zhijun Chen

As the basic support for the intelligent and connected transportation system, the intelligent and connected environment has become a research hot spot of common concern for universities, research institutes, enterprises, and government in the field of transportation. As a product of the deep integration of modern high technology with the transport industry, the automotive industry, and the information industry in an intelligent and connected environment, the autonomous driving map is also the core technology for the development and application promotion of connected and automated vehicles. Although there have been many studies on ICTSs in recent years, the construction of autonomous driving maps is still in its infancy.

The construction and development of autonomous driving maps in an intelligent and connected environment are of great significance for the progress and landing application of autonomous driving technology, which is mainly reflected in the following points:

First, autonomous driving maps can optimize not only for a single vehicle but also the entire traffic system.

Second, the development of autonomous driving maps can promote the large-scale industrialization of autonomous driving technology.

Finally, autonomous driving maps under an intelligent and connected environment can start with the overall system to provide safer and more effective autonomous driving travel services for travelers of different travel modes and technical levels.

This book focuses on the key technology of automatic driving map construction in an intelligent and connected environment. It aims to improve the stability of autonomous driving of connected and automated vehicles, promote the development of ICTSs, and provide readers with a knowledge system of autonomous driving map construction technologies in an intelligent and connected environment.

This book can be used by teachers and students of colleges and universities, researchers of scientific research institutes, technical personnel of enterprises and institutions, and government administrators engaged in traffic planning, traffic engineering, traffic control, and traffic management.

Limited by the author’s level and ability, this book inevitably has some inadequacies, and I sincerely hope to get readers’ criticism and correction.

1

Introduction

Abstarct

This chapter describes the development of the intelligent and connected environment and connected and automated vehicles; introduces the development status from road vehicle sensor target perception to vehicle motion recognition methods and prediction technologies; discusses the connotation of autonomous driving maps; briefly analyzes the problems encountered in the development of autonomous driving maps; and gives corresponding suggestions.

Keywords

Intelligent and connected environment; connected and automated vehicles; target perception; vehicle motion recognition; autonomous driving maps; vehicle motion prediction technology

1.1 Intelligent networked environment and intelligent networked vehicles

1.1.1 History of intelligent connected environment

With the rapid development of intelligent technology, a new generation of perception technology, artificial intelligence (AI) technology, communication technology, mobile internet technology, and the concept of intelligent networking have emerged. Intelligence refers to the intelligence of machinery and equipment, with traits similar to human perception, analysis, decision-making, control execution, memory, and other capabilities. Networking is the process of creating a network by interconnecting related equipment, personnel, institutions, etc. These nodes can be regarded as information nodes in the network, and there can be information interaction between the network and each node. The continuous iteration of new technologies, new concepts, and new models has prompted the comprehensive upgrade of intelligent networks in terms of perception, storage, sharing, interaction, and integrated services, and the connotation of intelligent network connections has also constantly enriched and improved. The intelligent networked environment refers to the real-time collection of various required information through various devices and technologies, such as sensors and global positioning systems, the ubiquitous connection between things and people, and the realization of intelligent perception, recognition, and management of things or people. Nowadays, the intelligent networked environment has penetrated the fields of industry, agriculture, environment, transportation, logistics, security, and other fields, effectively promoting the intelligent development of these fields, making the limited resources more rationally used and allocated, and improving the efficiency and effectiveness of the industry.

Among the wide application fields of intelligent networking, the transportation field is the most closely integrated with the intelligent networked environment. With the increase in car ownership, traffic congestion is becoming more and more serious, traffic accidents are emerging one after another, and a new traffic mode is urgently needed to solve a series of traffic problems. Intelligent network connections provide an opportunity for the exploration and development of new models. For example, real-time monitoring of road traffic conditions ensures that information is transmitted to drivers in time so that they can make timely adjustments and effectively relieve traffic pressure. The automatic road toll collection system is set up at the expressway intersection, which eliminates the time of picking up and returning the card at the entrance and exit, greatly improving the traffic efficiency of vehicles. At present, the intelligent transportation system is still in the exploration stage, and the development of intelligent network-related technologies will inevitably promote the maturity of the intelligent transportation system, which means that the problems of traffic congestion and traffic accidents will be better solved.

1.1.2 Composition and characteristics of intelligent network environment

The basic composition and characteristics of the intelligent networked environment are intelligence and networking.

Intelligence is the ability to be similar to human abilities, which can be categorized into intelligent perception recognition, intelligent analysis and decision-making, intelligent control execution, and intelligent memory learning.

Intelligent perception recognition is the ability to perceive one’s own state and environmental influence using sensor monitoring technology (pressure, temperature, position, flow, and other traditional monitoring technologies, image recognition, voice recognition, and other technologies), Internet of Things monitoring technology, Internet information monitoring technology, and social network information monitoring technology, with a highly reliable manual regular inspection method, real-time collection, analysis, upload of work parameters, operating status, environmental conditions, usage, public opinion and other aspects of information, for further intelligent analysis and decision-making support. For special equipment, emphasis should be placed on the perception and identification of faults, damages, failures, and hazard-related safety state parameters.

Intelligence analysis decision-making refers to the ability to analyze judgments and decisions based on perceptual information. The content of intelligent analysis includes the analysis of the quality level, safety status, use and management of individual equipment, comprehensive analysis of enterprises, institutions, regions, industries, quality, management, safety, and other aspects, and making faults, production stoppages, maintenance, testing, supervision and inspection, and scrapping decisions according to the analysis results. Through the analysis of a large number of model algorithms and a powerful knowledge base, an analysis and decision-making system based on big data is established, bringing together big data throughout the life cycle, constructing various types of analysis, diagnosis, evaluation, prediction, and early warning models, and then using big data mining, cloud computing, modern statistics, and other tools to analyze and make corresponding decisions.

Intelligent control execution refers to the ability to automatically respond to decision instructions. According to safety requirements, monitoring abnormal situations (such as equipment failure, damage, failure, accident, illegal operation, and lack of management), taking corresponding measures, and dealing with them in time; developing a remote automatic monitoring system, timely adjusting the excess process parameters, and correcting illegal operations in time. When there is a dangerous situation in the external environment, the safety protection device is activated in time to avoid any accidents or reduce the consequences as much as possible.

Intelligent memory learning refers to the ability to accumulate experience, learn, and grow; use cloud storage, modern databases, and other technologies to establish a big data platform; conduct data analysis and mining; use the accumulated data to guide the corresponding business and service work; continuously accumulate new empirical data in business and service work, and incorporate these new data into the database; constantly circulate and increase the quantity and quality of collected data; enrich the data knowledge base; and achieve self-learning and growth.

Networking means that equipment, personnel, institutions, and environments that make up the network are interconnected and connected with the node where the information is joined. Information can be exchanged between the intelligent network and the various nodes: the device can transmit information to the intelligent network, and the intelligent network can also distribute information to the device. Through the Internet of Things, mobile Internet, local area networks, and other modern information technology, the network realizes the intelligence of special equipment. A special equipment is based on multidimensional data, such as time and space, through modeling, construction, big data mining and other technologies, and various application development, finally making the entire network have perception recognition, analysis and decision-making, control execution, memory learning, and other functions to achieve network intelligence.

1.1.3 Connotation of intelligent networked vehicles

Intelligent connected vehicles can be defined as next-generation vehicles equipped with advanced sensors, controllers, and actuators, with intelligent and collaborative driving capabilities to ensure safety, comfort, and energy efficiency, reducing the burden on human drivers. In general, intelligent networked vehicles can be classified according to the level of driving intelligence. Europe, the United States, and China have published different classification standards for intelligent and connected vehicles. In 2014, the Society of Automotive Engineers published the Smart Car Level Standard, which is used by many companies around the world, such as Toyota, Nissan, Tesla, and Audi, to guide their research and development, which is summarized as follows:

Level 0, no automation, requires the driver to perform all dynamic driving tasks;

Level 1, driver assistance, longitudinal or lateral control when activated;

Level 2, partially automated, vertical, and lateral control when activated;

Level 3, conditional automation, which can provide part-time or driving mode-related performance to perform all dynamic driving tasks;

Level 4, high degree of automation, can perform all dynamic driving tasks even if the driver does not respond to the system’s request to intervene in certain driving modes or geographical areas;

Level 5, fully automated, once programmed with destinations, all dynamic driving tasks can be completed in at least all environments that can be managed by a human driver.

Compared with traditional smart cars that rely on onboard sensing systems and information terminals, intelligent networked vehicles emphasize the concept of integrating modern communication and network technologies. Vehicle-mounted sensing systems can play a role in scenarios with a short line-of-sight range and short reaction time, while modern communication and network technology have more advantages in situations beyond the line-of-sight range and long response time. Road, vehicle-person, vehicle-background communication, real-time and reliable acquisition of traffic environment information around the vehicle, and vehicle decision-making information can detect potentially related vehicles and road condition information in a wide range and accordingly plan and change driving routes, etc. Vehicle information interaction and fusion among various traffic participants, such as vehicles and vehicle-road, can play a good complementary role with the vehicle-mounted sensor system and can form collaborative decision-making and control, together constituting the cornerstone of intelligent networked vehicles.

1.1.4 Development status of intelligent connected vehicles

Intelligent networked vehicles rely on intelligent networked environments to develop rapidly. Several countries have issued corresponding policies to encourage the development of intelligent and connected vehicles. In March 2020, the US Department of Transportation (US DOT) released the Intelligent Transportation Systems (ITS) Strategic Plan 2020–25 (hereinafter referred to as the ITS Strategy), which clarified the vision of accelerating the application of ITS and transforming the way society operates and its mission of leading collaborative and innovative research, development, and implementation of intelligent transportation systems to provide safety and mobility for people commuting and cargo transportation, describing the key tasks and safeguards for the development of intelligent transportation in the United States in the next 5 years. In 2019, China issued the Outline for the Construction of a Powerful Transportation Country. It is pointed out that it is necessary to build a safe, convenient, efficient, green, and economical modern comprehensive transportation system. In December 2020, the European Union published its Sustainable and Intelligent Transport Strategy, which states that, when it comes to intelligence, innovation and digitalization will shape the modes of transport for passengers and goods in the future, promoting innovation and the use of data and AI for smarter transport.

Companies around the world are also accelerating the development of intelligent connected vehicles. According to the road test data of American companies in 2019, the autonomous driving system companies represented by Waymo and GM continue to lead in the field of road testing, with a total of more than 3.5 million miles of road testing (0.076 times per 1000 mi; takeover every 13,219 mi). For the sake of safety and technological advancement, commercial vehicle companies represented by Tesla and Daimler focus on open-source platforms and truck digitization research, launch big data platforms, abandon L3, and go straight to L4. Cruise ran 830,000 mi throughout 2019. In the first half of the year, Cruise had 43 takeovers (one takeover every 7635 mi), with a total test mileage of 328,000 mi. The second half saw just 25 takeovers (one takeover every 20,110 mi), for a total of 502,000 mi tested. Obviously, Cruise has made great progress, and the takeover rate in the second half of the year is even better than Waymo. After 8 years of research and development, Baidu Apollo has a test fleet of 500 vehicles, 2,900 patents for intelligent driving, 244 test licenses, and more than 12 million kilometers of self-driving road tests.

1.1.5 Development status of intelligent transportation system

The development of intelligent transportation systems is also inseparable from the intelligent networked environment. The key technologies involved in intelligent networked vehicles, including environmental perception technology, V2X communication technology, intelligent decision-making technology, information security technology, high-precision map, and high-precision positioning technology, have also received more research and attention. In terms of basic research, relevant research institutions at home and abroad have begun to actively research and explore scientific theories and key technologies in intelligent transportation systems, and have made fruitful progress in the intelligent transportation system architecture, resource management strategies, and access node selection mechanisms, laying the foundation for the future development of intelligent transportation. In terms of network architecture design, some researchers have introduced the concepts of software defined networks (SDN) and network function virtualization into the architecture design of the Internet of vehicles, providing an effective solution. Ku et al. give the concept of an SDN-based vehicle ad hoc network, analyze the feasibility of integrating SDN into vehicle ad hoc network architecture, and discuss its potential advantages and services in detail. Zheng et al. proposed a software-defined heterogeneous vehicle network architecture that is well-compatible with a variety of wireless communication protocols and, at the same time, ensures the real-time requirements of autonomous vehicle communication. This architecture provides a hierarchical control structure. The main controller grasps the global information and is responsible for the macrolevel resource allocation task, and the secondary controller focuses on the resource allocation strategy of each autonomous vehicle. Fontes et al. enumerated the development potential of software-defined in-vehicle networks, analyzed the challenges and difficulties of applying the existing SDN architecture to intelligent transportation systems from the perspectives of theory and practice, and verified the proposed framework based on Mininet combined with actual video streaming services, proving its feasibility. To ensure the driving safety of autonomous vehicles in the future intelligent transportation system and meet their needs for ultra-low latency and ultra-high reliability of the network, some researchers have introduced mobile cloud computing and mobile edge computing into the design of network architecture, and satisfactory research results have been