Large cities across the globe face serious issues related to traffic, largely due to population growth and the resulting increase in the number of vehicles on the road. In this article we address some examples how the integration of Artificial Intelligence (AI) with modern Intelligent Transportation Systems (ITS) solutions is making it possible to predict and control traffic flow with remarkable accuracy, therefore optimizing efficiencies, promoting driver safety and increasing sustainability.

ITS Basics

The basic engineering behind an Intelligent Transportation System consists of electronics, communications, and information processing which are utilized individually or in combination. IP cameras, sensors, connected traffic light systems, smart toll gates, edge devices, and IR or microwave technology are integrated into ITS to gather traffic data. Transit data collected from these roadside devices is then routed to a remote Traffic Management Center (TMC), either through a cabled infrastructure, or through a wireless system (cellular, radio, microwave). The TMC acts as the nerve center of an ITS, and is where the connected devices, video monitoring, servers, network infrastructure and control systems work together to provide actionable insights for more effective traffic management, incident response, and information sharing. This process is increasingly being aided by AI and its subsets of Machine Learning and Deep Learning.

Antaira's Industrial Networking Solutions for ITS

Antaira's industrial networking solutions converge and support the broad and diverse technologies that make up modern ITS systems. This includes AI - enabling faster throughput of data and more flexible connectivity by incorporating Ethernet, wireless, fiber, PoE, serial and other network interfaces. Our high-port count Gigabit managed industrial Ethernet switches, gateways, and secure industrial wireless routers have long been relied upon by ITS professionals around the world, not only to connect remote machines, sensors and cameras in harsh environments, but to power them on roads, bridges, and tunnels with PoE, saving money and time.

Traditional Traffic Forecasting

Decisions regarding traffic forecasting and management of industrial networks are the most important responsibilities of an ITS. Analysis of data and decision-making have traditionally been made by humans, more specifically, the engineers in the TMC. Like all human decisions, they can be prone to bias, inattention, fatigue or lack of proper training, among other factors. By mimicking human reasoning while processing volumes of data far faster than even a few human mind or large team of engineers, artificial intelligence is rapidly transforming forecasting of industrial networks.

In the past, time series analysis and other parametric models based on historical data were used to forecast traffic flow. Observations at different road locations were manually recorded in a time series in an attempt to recognize temporal patterns. For estimating short-term traffic flow, the Auto-Regressive Integrated Moving Average (ARIMA) model is a well-known and accepted paradigm. Another model is the Kalman Filtering method. With the rise of AI and machine and deep learning however, these manual systems are becoming automated.

Applying AI Algorithms to Traffic Forecasting

Artificial Intelligence, along with its subsets of Machine Learning and Deep Learning, have shown humans the capacity to process and analyze enormous volumes of real-time and historical data. This data is utilized by machines to create statistical models for classifications in developing highly predictive traffic algorithms. AI processes replicate human intellect and reasoning through the application of rules to arrive at a precise set of findings.

When artificial intelligence is applied to ITS, AI predictive algorithms enhance traffic flow forecasting models by accurately and quickly evaluating the vast amounts of complex data acquired by loop detectors, ultrasonic wave detectors, pedestrian sensors, and cameras. This input data is synthesized through data fusion to obtain useful traffic information such as speed, trip durations, and congestion. From the chaos of process data, patterns emerge that enhance driver safety, reduce congestion, promote roadway efficiency, and improve emergency response and transit system operations. By correcting foreseen traffic flows, AI can mitigate the risk of traffic accidents that take the lives of 42,000 Americans every year by largely eliminating several of the human factors that are a frequent cause.

Along with forecasting, the data of the industry also sharpens algorithms through supervised, unsupervised and reinforced learning. Reinforced learning has proven to be very useful in ITS. It occurs when the system detects changes in the surrounding environment, prompting an automated action which is assigned either a positive or negative reward, depending on the result. The system seeks to maximize positive rewards, thus learning to “do the right thing” from its decisions based on its training data and past actions.

Machine learning tools and deep learning techniques can not only improve traffic flow on existing roads but also assist in the design of new, more efficient road construction, literally paving the way to Smart Cities.

1. Use Cases for Adaptive Traffic Signal Control

According to a Forbes article, the transportation analytics firm INRIX stated early AI research that the average U.S. motorist spent 51 hours sitting in traffic last year — 15 hours more than in 2021. Traffic congestion at busy intersections wastes productive time, stresses both drivers and pedestrians, increases air pollution, and leads to higher traffic violations and accidents.

Conventional fixed-time signal control systems lack the ability to adapt to varying traffic congestion levels and patterns. Fixed-time control systems operate on a time-of-day sequence, meaning that predetermined timing plans go into effect during certain times of the day. This “set it and forget it” approach has proven to be woefully inadequate. Roads become so congested that cars move at a snail's pace towards the intersection, waiting for several complete light cycles.

Adaptive Signal Control Technology (ASCT) leveraging sensors and AI algorithms significantly improves traffic performance metrics at intersections, according to numerous published studies. According to an article from the US Department of Transportation states that in areas of rapidly changing or unpredictable traffic volumes, adaptive signal timing control may improve system performance by 5 to 30 percent. Unfortunately, in the United States less than one percent of existing traffic signals use ASCT.

As its name implies, ASCT automatically adapts signal timing in response to unexpected changes in traffic patterns. ASCT essentially makes traffic signal operations proactive by accommodating emerging patterns by adjusting the time green lights start and end. Available adaptive signal control tools and support technologies today include the Split Cycle Offset Optimization Technique (SCOOT), Sydney Coordinated Adaptive Traffic System (SCATS), Real Time Hierarchical Optimized Distributed Effective System (RHODES), and Optimized Policies for Adaptive Control (OPAC) "Virtual Fixed Cycle" and ACS Lite.

Like all ITS and AI systems, data is at the heart of ASCT. It is collected by either video cameras or sensors like inductive wire loops embedded in the pavement, or a combination of both, at or near the intersection. Cameras can better identify types of vehicles, as well as detect pedestrians and bicycles, making them a better choice for AI applications. ASCT algorithms train on large datasets of massive amounts of images to learn how to identify and classify objects within an image. For instance, it can distinguish a bicycle from a motorcycle. Reinforced Learning, mentioned earlier, is showing promising results in ASCT.

Unlike human beings, AI is capable of tracking hundreds of cameras at once. It is also far superior to human experts at accurately counting autonomous vehicles already at the intersection, calculating gaps between the vehicles, and simultaneously predicting trajectory of autonomous vehicles and all road users at the intersection (vehicles, pedestrians, public buses, bicycles). AI can determine behaviors before they are made, and evaluates data in seconds and modifies signal timing that the ASCT implements. The ASCT process is continuously updated to ensure the signal timing is responding to real-time conditions at the intersection.

Antaira's Technology in Adaptive Traffic Signal Systems

To fulfill real-time ASCT applications, reliable end-to-end network dataflow is crucial. Antaira’s industrial PoE switches deliver power to the sensors, machines, networks and cameras collecting data for processing. In addition, many PoE switches feature fiber optic ports to enable faster, interference-free data routing back to the traffic management center over long distances.

2. Use Cases for Dynamic Route Guidance and Traffic Management

AI-enabled algorithms can determine the most effective and efficient routes for a range of transportation and logistical operations by evaluating enormous volumes of data, historical patterns, and real-time information in a process known as Dynamic Route Guidance (DRG). The upshot is an optimized route that minimizes travel time and distance, reducing fuel costs and environmental impact. The term “dynamic” refers to AI’s ability to adapt the suggested route in real-time in response to events such as emerging traffic congestion, accidents, public events, or weather, as the motorist is driving. DRG applies Machine Learning to continuously improve the navigation algorithm using new data and analyzing past experiences.

DRG is a convenience for commuters but is essential to commercial delivery services in meeting their goal of maximizing fleet efficiency. Given the number of orders, fleet size, load capacity of different vehicles, and other logistics variables involved in moving goods from one place to another. DRG is far beyond the capacity of human dispatchers. For delivery services leveraging DRG, different dynamic routes are built daily for each driver since customer orders will change from day to day. By monitoring conditions, DRG can update delivery drivers on an ongoing basis, allowing the business to provide customers with more accurate ETAs.

DRG follows a series of steps, all of which require interconnected network communications. First, DRG collects data on the places that need to be visited, the distances between the stops, the vehicle's load capacity, the time allotted, historical traffic statistics, traffic lights, speed limits, the weather, and other pertinent logistics. Next, it selects the ideal navigation algorithm which could be a heuristic algorithm, a metaheuristic, or a combination of multiple algorithms. An initial route is generated, either randomly or based on heuristics, that serves as the starting point for optimization. Artificial intelligence then explores optional routes, swaps the order for scheduled stops, and evaluates these changes based on user-defined criteria and objectives until a termination condition is met - for instance, a time limit or a set quality level. Being dynamic, DRG sends and receives real-time data updates on traffic bottlenecks, road closures or changing weather conditions, and will alter the driver’s route for the best results possible.

Antaira’s Role in Dynamic Router Guidance

Even if you are not in the logistics business, it is easy to see the benefits of dynamic route guidance. What is less obvious are the complexities of its network architecture. Roadside sensors and cameras must continuously send data to traffic control systems where the raw data then is analyzed by algorithms to determine the optimal route and then have the route distributed to mobile phones, the Internet and car navigation systems, all in real-time. Antaira’s industrial networking equipment makes this connectivity possible. For instance, our extensive portfolio of managed and unmanaged industrial switches reliably interconnect DRG devices with industrial networks and systems. Antaira also offers industrial serial device servers equipped with RS232/422/485 connections to bridge the gap between legacy serial sensors and the control center by using Virtual COM and a TCP/UDP socket function.

3. Use Cases for Incident Detection and Prevention

Artificial intelligence can identify and detect high level of traffic incidences ranging from collisions and wrong way drivers to broken down vehicles and police roadblocks. Once an incident has been identified, this information is used to quickly assign staff to the area and guarantee the appropriate response. Additionally, the data can be utilized with human intelligence to create and speed up measures like the control applications diverting traffic away from the incident by broadcasting it on digital signage.

As a result of AI integration, how transportation organizations handle traffic incidents is rapidly changing. Artificial intelligence can evaluate real-time traffic data to forecast incidents before they happen, suggest mitigation measures, and improve traffic flow. AI can be taught to recognize different classes of incidences, as well as specific tasks. It only needs to be fed the images for AI algorithms to be trained to distinguish, for instance, dangerous objects dropped from a passing truck from supplies being used by a maintenance crew for road repairs.

The best way for drivers to survive an accident is to not have one in the first place. AI forecasting can calculate the likelihood of an accident occurring on a specific stretch of highway at an exact time of day. Transportation authorities can use this data to lessen the prospect of primary accidents, stop the occurrence of secondary accidents, and increase the safety of first responders if an accident does occur. One preventive tactic frequently employed is to assign a police officer to the targeted road area at the time AI predicted accidents occur most often, encouraging motorists in the area to drive slower and follow traffic regulations.

Artificial Intelligence is the Future of Traffic Management

There’s a great future ahead for AI in ITS. In this article we touched on examples of traffic forecasting, automated intersections, route guidance and incident response. Other applications for AI in ITS are emerging daily, from smarter parking to better law enforcement support.

Artificial intelligence is making its way to the ITS network edge, just as it has in the automotive industry, enterprise and Industrial IoT (IIoT) arena. Consider the new generation of AI-based roadside cameras featuring embedded intelligence, meaning that it’s no longer necessary to send camera data over the network to a central server or cloud service for processing. As a result, the network is not overloaded when there is no detection, and detection can happen with much less latency. ITS is also moving to the sky with the use of AI equipped camera drones monitoring highways and remote roads.

With AI poised to reshape transportation, the time is right to partner with Antaira on your next ITS project. At Antaira, the future of ITS is now. Our team of ITS technical experts are experienced with AI, machine learning, deep learning models, sensors, IP cameras, roadside systems, and real world ITS applications. Contact Antaira today at (714) 671-9000 to speak to a live representative, send us an email at ITS@antaira.com, or visit our ITS webpage (www.antaira.com/its).