AI Applications in Telematics, Traffic Prediction, and Transportation Planning

Regardless of where we reside, encountering traffic jams and incidents on the road is a common occurrence, often a daily reality for many. While enduring a 30-minute delay due to accidents or recurring congestion may be common for urban commuters, even a few minutes of gridlock in rush hour can be a dreaded experience for drivers in less densely populated areas. Traffic congestion has become so pervasive that it is now an integral part of our daily lives.

Increasing roadway capacity requires substantial funding and time, but advancements in computational, communication, sensing, and vision technologies within transportation engineering have been instrumental in enhancing safety and traffic efficiency and providing drivers with real-time and predictive traffic information.

Iteris: At the Forefront of Telematics and Real-time Traffic Updates

For decades, Iteris has been a trailblazer in providing advanced technologies and traffic planning services to cities, states, and Metropolitan Planning Organizations (MPOs). However, lesser known is Iteris' provision of telematics services directly to motorists. Through collaborations with industry leaders managing communication and digital data dissemination infrastructure, Iteris delivers real-time traffic updates to vehicles across North America via satellites, Digital Audio Broadcast (DAB), or GSM cellular networks.

To provide minute-by-minute real-time traffic flow information covering 3.8 million miles of road in North America, Iteris' ClearData™ engine is robust enough to process billions of GPS data points daily. Additionally, to offer comprehensive traffic condition updates to travelers, our ClearData engine executes hundreds of software programs per minute to gather real-time data on accidents, construction, and work zones from over 150 public and private sources.

This is where artificial intelligence (AI) plays a pivotal role in extracting and integrating every piece of information into the ClearData engine.

Standards in Traffic Data Communication

In order to standardize the delivery of traffic and traveler information to head units sitting inside vehicles manufactured by various Original Equipment Manufacturers (OEMs) globally, specifications such as RDS-TMC (Radio Data System - Traffic Message Channel) Alert-C (ISO 14819) have been established to govern the structure and presentation of traveler information.

The RDS-TMC Alert-C standard defines how traffic flow and incident data, along with their respective times and locations, are encoded into bits and bytes to facilitate machine-to-machine communication while utilizing minimal bandwidth. While traffic flow data naturally lends itself to numerical representation, incident descriptions are presented in text format. To deliver traffic incident updates to vehicles according to the RDS-TMC Alert-C standard, these incidents must be converted into a numerical format in both location and description domains.

Figure 1: Real-Time Traffic Information for In-Vehicle Navigation

Expert Systems Versus AI in Traffic Information Processing

The Alert-C event code table plays a crucial role in converting incident descriptions into numerical values, much like how TMC location codes aid in identifying traffic event locations for commercial maps like HERE maps and TomTom maps.

However, ensuring the accuracy of incident description to Alert-C event code conversion in real-time presents a significant challenge. With over 1,400 entries, this table serves as a comprehensive reference for categorizing various incident descriptions. Below are a few examples.

Figure 2: A Snip of Alert-C Event Code Table

The description "bridge closed" in the traffic event below was translated into the Alert-C event code "24" before its transmission to vehicles for processing by the head unit.

"In West Bloomfield, bridge closed on Putnam Dr SB between Lawndale Ave and Belgrave Ave."

The traffic event below was automatically converted to Alert-C event code "1559," closed due to parade, without human intervention.

"49TH ANNUAL BROOKLYN ST. PATRICK S PARADE. THIS STREET WILL BE CLOSED FOR THE 49TH ANNUAL BROOKLYN ST. PATRICK S PARADE ON SUNDAY, MARCH 17TH, 2024 AT THE DISCRETION OF THE NYPD IN BROOKLYN."

AI-Driven Approaches to Incident Data Categorization

The automation process demonstrated above is executed in real time through two key steps:

1. Utilization of regular expressions to extract crucial traffic information from the free-form text descriptions provided by data providers.
2. Development of an Expert System comprising over 70,000 rules, which leverages archived data samples, to automatically match incident descriptions into Alert-C event code.

This systematic approach ensures efficient and accurate processing of incident data, enabling swift dissemination of standardized traffic information to vehicles and enhancing overall road safety and efficiency.

Nonetheless, our ClearData engine operates around the clock, continuously capturing new incidents and maintaining a repository of over 20,000 incidents at any given time. Hence, managing and updating the extensive set of rules within the Expert System poses challenges over time. To address this issue, an AI model has been developed.

From an AI perspective, this challenge is tackled using two distinct approaches:

1. Categorization Approach (CA): In this approach, the AI model directly identifies the hidden relationship between incident descriptions and Alert-C event codes.
2. Natural Language Processing (NLP) Approach: Here, the AI model identifies the closest Alert-C event message based on the incident description and then maps it to the corresponding Alert-C event code. This approach mirrors the methodology employed by Iteris operators when the Expert System encounters difficulties in finding the correct match due to the ambiguity of incident description.

A dataset comprising 16 million archived incidents was utilized to construct an AI deep learning model consisting of 12 layers and encompassing 20 million trainable parameters. Within this dataset, 80% of incidents were allocated for model training, while the remaining 20% were reserved for validation.

This innovative use of AI not only enhances the efficiency and accuracy of incident-to-event code mapping but also demonstrates the potential for AI to optimize complex systems in real-time applications.

AI-Driven Solutions for Traffic Prediction and Planning

The recent surge in AI adoption marking a significant evolution in the realization of its potential is driven by several key factors:

1. Advancements in computing technologies and resources: The availability of powerful computing technologies and abundant resources, coupled with increasingly affordable cloud storage, has democratized access to AI infrastructure.
2. Expansion of algorithms: Computer and data scientists have developed brilliant algorithms that not only transcend traditional text and image processing but also extend to audio and video analysis, broadening the scope of AI applications.
3. Digital data explosion: The proliferation of digitized multimedia data since the advent of the internet in the 1990s has provided a vast resource for training AI models on an unprecedented scale.

Having served the transportation engineering and telematics industry for decades, Iteris has amassed a wealth of SPaT (Signal Phasing and Timing) data for arterials, along with incident, flow, and GPS probe-derived information spanning the entire country. By applying geospatial and temporal tagging to these datasets, a standardized traffic information data lake can be established, facilitating AI model training, and enabling the creation of diverse applications.

1. Incident impact delay time prediction: Frequently, real-time driving speed (flow) and traffic incidents are presented as distinct data elements on mobile devices or within vehicle-integrated traffic applications. At the request of OEMs, Iteris developed a statistical model to estimate travel delay time attributable to traffic incidents. By harnessing AI's capacity to unveil the intricate and concealed relationship between flow and incidents across spatial and temporal domains more efficiently, a multitude of applications can be rapidly developed to address the mobility challenges of today.
2. Incident hot spot prediction for transportation planning: The concept of compiling archived incidents into an incident hot spot database and potentially correlating it with traffic volume counts for transportation planning and traffic operations has been in practice for several years. However, the standalone use of an incident hot spot database has limitations, and obtaining traffic volume counts can be both expensive and time-consuming. Integrating AI to uncover the relationship between incidents and the accumulation of probe counts and probe paths derived from GPS data offers a comprehensive and cost-effective means for traffic planning and incident mitigation efforts. Furthermore, with a constant influx of billions of GPS probes daily, potential traffic issues can be identified early, allowing for the implementation of mitigation strategies to prevent them from escalating.
3. Short-term traffic flow prediction: Traffic flow prediction is indispensable for navigation and trip planning, benefiting daily commuters and logistical operations like fleet management and timely deliveries. Traditional navigation applications typically utilize simple statistical models, relying on instantaneous speed around the traveler and historical traffic patterns at the destination to estimate trip travel time. Iteris developed an exploratory traffic prediction model to support OEM's routing applications. Applying AI models can further enable dynamic interactions between real-time and historical traffic patterns throughout the entire journey, offering all travelers a more accurate depiction of overall traffic conditions. This approach provides a more realistic view of traffic conditions, enhancing navigation and trip planning experiences for users across various scenarios.
4. Synthetic data generation for traffic planning and simulation: Creating AI models entails significant investment in time, resources, and skilled personnel to gather, clean, and standardize vast amounts of data from diverse sources and contexts. While this process can be lengthy and costly, the rewards often extend well beyond initial objectives. Nowhere is this more evident than in transportation planning and traffic operations, where painstaking collection of ground truth data is essential for crafting implementation strategies and conducting cost-benefit analyses before and after strategy execution. Synthetic data generated by AI models offers an ideal solution to these challenges, providing valuable insights and reducing the resource burden associated with traditional data collection methods.

AI as a Catalyst for Future Mobility Solutions

These applications demonstrate the transformative potential of AI in addressing contemporary transportation challenges, paving the way for more efficient and data-driven mobility solutions. To learn more about Iteris' AI solutions, contact us to speak with an expert on our team.

About the Author:

Jerome Chen-Chan is Vice President of Traffic Analytics at Iteris. Connect with Jerome on LinkedIn.

Public link

Please use the above public link if you want to share this noodl on another website.