Artificial Intelligence Driving into "Expressway": Application Scenarios and Examples of Traffic Artificial Intelligence

Cui Weixue, China Traffic Informationization Reporter At present, a new generation of general technology industrial revolution is in the ascendant, which has brought subversive changes to digital and intelligent production and life. After a new wave of development, artificial intelligence has penetrated into all aspects of people's production and life and become the representative of a new generation of general technology. Under the "new infrastructure", new applications, new formats and new models built around "AI+" are constantly emerging, and artificial intelligence has given full play to the "head geese" effect.

As an important starting point of the wave of "new infrastructure", the field of intelligent transportation has attracted much attention. The development concept, management mode and service experience brought by digital technologies such as artificial intelligence, 5G and industrial Internet have ushered in a global "intellectual change". Nowadays, the intelligent construction of expressways everywhere is in full swing. With the national expressway officially entering the era of "one network" operation, the application of AI in expressway has entered a stage of rapid development. Based on the experience sharing in the 22nd China Expressway Informatization Conference, this paper briefly combs the integrated application of artificial intelligence in smart highways.

With the successful completion of station withdrawal, the national expressway has realized "one network" operation. In this situation, road section management units put forward new requirements for operation management: (1) They hope to make further breakthroughs in technology, service, management and coordination, and promote the realization of "efficient and economical" toll-free scenes with few people; (2) It is necessary to realize the fine management of road sections, collect all the tolls receivable and ensure their own rights and interests; (3) The massive and diverse road network equipment and facilities in the road section urgently need the support and guarantee of intelligent and automatic technology to ensure the safe and stable operation of the side facilities. In order to solve the above pain points, China Merchants Huasoft Information Co., Ltd. relies on the "AI+ Cloud" technology to build version 2.0 of smart charging, which fully empowers charging audit, operation analysis, operation and maintenance management, operation monitoring and basic charging services.

AI+ Cloud, A Breakthrough Attempt

China Merchants Huasoft has built a unified smart charging cloud platform to migrate the charging business and related transportation management business to the cloud, and efficiently realize multi-dimensional collaborative management among businesses. With the empowerment support of the platform, the intelligent lane robot is also used to create an unmanned toll station solution for road business units.

Unmanned toll station is a breakthrough attempt in the application of "AI+ Cloud" scenario, a pragmatic measure to reduce costs and increase efficiency in road sections, and an inevitable stage in the development of smart toll collection. The AI capability of the front-end lane robot is combined with the overall planning capability of the cloud smart toll cloud, which greatly enhances the service value of the current lane toll.

No one charges, showing IQ.

Lane robot is an intelligent front-end AI device of unmanned toll station. The whole machine integrates all kinds of hardware modules needed for lane toll collection, supplemented by human-computer interaction engineering design, and with the help of edge computing, intelligent voice, 4G/5G and other technical means, it realizes unmanned self-service toll collection and automatic handling of special situations.

The solution of unmanned toll station realizes intelligent front-end equipment, comprehensive on-site support, intelligent back-end system and economical operation investment. Compared with the traditional toll-free mode, it has the following outstanding capabilities: it is suitable for various lane application scenarios, such as ETC toll collection, MTC toll collection and mixed toll collection. , supporting various non-cash payment methods such as ETC card, WeChat, Alipay and China UnionPay Express, and will have the ability to accept cash payment in the future; The integrated artificial intelligence unit supports intelligent voice interaction with car owners, which can quickly locate user problems without on-site intervention and provide convenient and effective customer service for car owners; At the same time, it can greatly reduce the labor cost of the road section management unit.

Demonstration application, accelerate landing

At present, China Merchants Huasoft Smart Toll Cloud Platform has been applied in Guilin Company of China Merchants Expressway and its surrounding sections, and the operation effect is good, which basically meets the needs of daily toll collection, monitoring, inspection and analysis, operation and maintenance, and greatly improves the transportation and management efficiency of sections. The toll plaza of the northbound exit of Xilong Toll Station on the First Ring Road in Foshan has opened a demonstration lane for unmanned robots. Lane robots in all lanes of Harbin-Dalian Expressway have also been installed and put into use.

In the practical application of lane charging, 95% of the charging business can be completed by the intelligent logic processing ability of lane robot. For less than 1% of the problems that need to be handled on site, services can be provided through the infantry combat tool "integrated service feedback control terminal".

2019165438+1013 The General Office of the Ministry of Transport issued the National Expressway Video Networking Monitoring Work Implementation Plan and the National Expressway Video Networking Technical Requirements, proposing to speed up the construction of a "visible, measurable, controllable and serviceable" expressway operation monitoring system. Video surveillance has played an active role in improving the efficiency of operation and management and the ability of public service. With the continuous accumulation of massive video data, how to realize real-time detection, dynamic monitoring, intelligent control, timely service and accurate prediction of intelligent monitoring has become an important challenge in the current technical field. Dr. Zhang Hailiang from Shanxi Communications Vocational and Technical College shared the application of AI technology in expressway video cloud networking in combination with the construction of expressway video networking in Shanxi.

Study hard and show your talents.

From the perspective of architecture, expressway video networking adopts cloud, edge and end three-tier architecture. Among them, the edge intelligent analysis system adopts a new generation of video traffic incident intelligent monitoring system, which has the characteristics of access at any time, real-time analysis, real-time alarm and high accuracy. With the continuous accumulation of data, the longer the system runs, the smarter the recognition algorithm and the higher the detection accuracy. Based on deep learning technology, the system can achieve the following functions.

Traffic incident and traffic flow detection: By using the multi-target detection and target tracking algorithm based on deep learning technology and accessing the video data of high-speed video cloud networking, the edge intelligent analysis system can realize traffic incidents such as road congestion, traffic accidents, vehicle retrograde, illegal parking, pedestrian intrusion, throwing articles, lane change, construction, fireworks, fog, emergency occupation, and traffic conditions such as traffic flow and traffic parameters. Vehicle structure analysis: Vehicle feature data is extracted by intelligent algorithm to realize vehicle structure analysis, which is applied to vehicle judgment, violation punishment, fee evasion detection, fee inspection and other businesses. Video quality diagnosis: Through image recognition algorithm and video picture quality diagnosis and analysis, patrol and evaluate the status of equipment in the field, find equipment problems in time and respond quickly.

Data analysis and efficient management

Through "AI+ Big Data" technology, the provincial cloud platform integrates the video data of road sections and the massive perceptual data of the edge intelligent analysis system, and carries out data calculation, data analysis, data mining and comprehensive judgment through massive data model training and deep learning to realize intelligent monitoring applications such as intelligent supervision, traffic situation analysis, forecasting and early warning, and emergency response. At the same time, it provides data catalog, API, data application and visual display to external systems and applications through data portal.

Big data analysis after expressway video networking can not only realize real-time monitoring, prediction and early warning of industry operation situation, but also provide scientific basis for industry operation management decision-making, and provide data support for function and business process supervision, decentralized domain management and visual business display, thus improving operation management efficiency. In terms of logical architecture, the provincial center deployed an AI-based provincial intelligent early warning platform and a provincial video cloud platform. Based on the front-end information collection terminal equipment, the application platform of road video uploading and video big data intelligent analysis, the fusion application of cloud-connected video data is realized.

Shanxi pilot has achieved remarkable results.

In recent years, Shanxi Expressway has been actively carrying out research on expressway intelligent operation monitoring, especially in the intelligent analysis system and platform of expressway video networking. The intelligent platform based on AI has achieved ideal results in specific practical applications. The system platform construction takes the most advanced high-performance GPU cluster as the physical carrier, and pioneered the distributed deep learning algorithm and multi-task neural network model, which greatly improved the accuracy and parallel efficiency of the system and made the system extremely advanced, which is embodied in the following four aspects.

(1) is well-informed. Through deep learning of massive training samples, and with the increase of system deployment and application, the accuracy of the system is getting higher and higher. (2) The platform is advanced. The system adopts Liu Yitong, a big data computing engine based on data stream, and uses a neural network distributed training platform, which improves the convergence speed, shortens the training time and improves the model iteration efficiency. (3) The algorithm is advanced, and a variety of algorithms have been developed, which can realize accurate identification of traffic events and vehicle information. (4) continuous evolution. In the specific system application process, after the system test results are manually confirmed, the positive and negative samples are constantly increasing, which can realize continuous learning and evolution.

Huang Lairong, a senior engineer of Fujian Expressway Information Technology Co., Ltd., shared the scheme of Fujian networked toll operation monitoring system based on artificial intelligence and Internet of Things. After the cancellation of provincial toll stations, the ETC toll display system has been optimized, and the overall network toll collection system in Fujian Province runs smoothly. Online billing is highly dependent on operation monitoring, which is mainly reflected in online billing, status list synchronization, minimum cost allocation of the whole network, portal billing module upgrade and so on. There are many difficulties, such as many points, many monitoring contents, complex equipment types and high requirements for operation monitoring. Therefore, it is necessary to have a set of systematic tools to quickly diagnose the problems of driveway, gantry and back-end system, so as to improve the efficiency and accuracy of system operation and maintenance. Fujian Expressway has integrated the original toll operation and maintenance management system, ETC lane operation monitoring system and ETC gantry operation monitoring system, and has successfully launched the Fujian Expressway networked toll operation monitoring system, ensuring the normal operation of all levels of networked toll systems.

Clear goals and complete functions.

The construction of the monitoring system of provincial networked toll operation mainly focuses on the following four goals: improving ETC customer service level; Guarantee transactions between units and provinces, and realize "subsection charging and unified export charging"; Promote manufacturers to improve product quality and after-sales service level; Improve the daily mechanical and electrical maintenance level.

The construction contents include: the operation monitoring of lane system, including lane equipment monitoring, lane industrial computer monitoring, lane data monitoring and lane transaction monitoring; Gantry system operation monitoring, including ETC gantry equipment monitoring, gantry host monitoring, gantry data monitoring, lane transaction monitoring; Back-end system operation monitoring includes back-end equipment monitoring, back-end host monitoring, back-end application monitoring and back-end data monitoring; System alarm, including classified alarm, serious alarm prompt and alarm push; Run monitoring tools, including system upgrade inspection tools, parameter distribution inspection tools and fault diagnosis inspection tools.

Intelligent analysis, scientific early warning

The system uses components such as NumPy, Pandas and scikits-learn based on machine learning, and can analyze the fault causes through decision tree regression algorithm. K-means clustering algorithm is used to find abnormal values, analyze and predict the health status of gantry or driveway; In order to use naive Bayesian algorithm to predict equipment failure, it is necessary to maintain the equipment in advance for subsequent intelligent analysis.

1. Manufacturer theme analysis classifies equipment manufacturers, and makes statistics on transaction success rate, capture rate, abnormal quantity and other data, so as to promote equipment manufacturers to provide high-quality products and improve after-sales service level. 2. User topic analysis (1) peer media status: It indicates that the OBU battery is low, locked, damaged or about to exceed the effective expectation. (2) Recharge reminder: when the stored value card is lower than the user's routine to form a certain proportion, it will be recharged. (3) Add status list reminder: remind users when they are listed in the status list. (4) Combination of formation rules and user services: According to the formation rules of users, provide information on road conditions and services (such as service areas) along the way. (5) Continuous abnormal reminder: When the OBU abnormal transaction on the driveway and gantry reaches a certain threshold, the user will be reminded and recalled for inspection. 3. Fault prediction and early warning (1) Lane system fault prediction: the decline of traffic efficiency and the increase of abnormal transaction ratio may indicate lane system fault; The decline of lane license plate recognition rate may indicate that license plate recognition has failed or needs maintenance and optimization. (2) Early warning of gantry system failure: an increase in the proportion of gantry abnormal transactions and a decrease in the capture rate usually indicate that the gantry system has failed; The decrease of RSU or license plate capture rate in a lane usually indicates that the equipment is out of order or needs timely maintenance. (3) Early warning of equipment and environment association: Through cluster analysis or correlation factor analysis, identify the regular relationship between equipment and environment, such as the relationship between power jump and thunderstorm weather, and the relationship between license plate recognition rate and weather.

There are some pain points in the tunnel, such as closed space, frequent accidents, difficult disposal and weak prevention and control. It is urgent to develop new tunnel risk prevention and control technologies and devices on the basis of existing technologies. Under the east wind of "new infrastructure", a set of intelligent traffic patrol system platform supporting dynamic cruise, giving consideration to high precision and real-time came into being. Associate Professor Ma of Chongqing Jiaotong University introduced the platform at the meeting, which can realize water seepage detection, crack detection and tunnel environment detection. Realize patrol inspection of traffic accidents, and detect, analyze and process traffic parameters such as traffic operation status, important traffic infrastructure, traffic volume and speed in real time. The detection accuracy is more than 80%.

Integrate innovation, improve quality and upgrade.

As the front-end equipment of the platform, the intelligent inspection robot of tunnel cloud integrates technologies such as artificial intelligence, 5G, virtual reality and industrial Internet of Things, upgrades the intelligent management level of tunnel by relying on high-end scientific and technological means, and actively responds to the national "new infrastructure" policy.

The intelligent inspection robot of tunnel cloud adopts edge artificial intelligence technology. Compared with the traditional cloud-based computing method, this technology has the advantages of low delay, high energy efficiency, privacy protection, reduced bandwidth occupation, timeliness and high environmental sensitivity, which makes the tunnel inspection robot more agile in perception and more intelligent in risk identification and emergency decision-making. 5G has the transmission capability of large bandwidth and low delay. The platform has established a regional high-capacity 5G transmission network based on edge devices, integrated infrared thermal imager, laser/millimeter wave radar, high-definition panoramic camera and other cutting-edge technologies, and explored the application demonstration of 5G network in highway tunnels.

Intelligent detection, safe and efficient

The intelligent inspection robot in the cloud helps the "new infrastructure" and "traffic power" to accelerate, and realizes all-weather and unattended intelligent inspection of roads, bridges and tunnels, which can ensure the safety of tunnels to the maximum extent. Compared with the traditional manual inspection, it has the following advantages:

(1) The electromechanical systems such as lighting, ventilation and fire fighting are networked into a whole through the intelligent robot in the cloud to realize adaptive joint control; (2) The robot is equipped with sensors such as lidar and thermal imager, which can dynamically sense disasters such as cracks and leakage; (3) The robot can arrive at the scene in the first time and conduct real-time remote traffic monitoring, emergency rescue and evacuation command.

A sudden epidemic in 2020 put forward very urgent requirements for "new infrastructure". The remote, non-contact and intelligent response to the epidemic has stimulated new market demand, and forced traditional industries to accelerate the pace of digital transformation. Therefore, the construction of intelligent transportation has become a realistic demand for urban development. As the main content of "new infrastructure", the deep integration and collision of new technologies, such as "big things move wisdom" represented by artificial intelligence, have formed the core competence of a new generation of information infrastructure. Traffic AI is the general trend. In addition to the contents introduced in this article, AI has also played an important role in urban public transportation, autonomous driving and other fields. In the field of smart high-speed, AI is fully empowered in the cloud, tube, edge and end, bringing new upgrades to application scenarios such as charging, audit monitoring, driving into high-speed, getting on the bridge and entering the tunnel, and it is everywhere. The iteration speed of science and technology is amazing, and the wave of 5G is coming quickly. The outbreak of traffic will promote the development of data processing and analysis capabilities, and artificial intelligence will also usher in new opportunities and challenges. With the rapid development of the new generation of information technology, all roads will be endowed with powerful subversive forces, leading to a boundless and unbounded intelligent future.

(The original text was published in China Traffic Informatization, No.3, 202 1).