Avoiding bumps in the road when designing AI-powered traffic management systems

Since its introduction in the 1950s, artificial intelligence (AI) software has transcended its theoretical existence in research labs to become omnipresent in our lives. This early AI research software has not only fueled an explosion in efficiency, but it has also opened the door to entirely new opportunities in business, education and government. Most of us use AI software daily in e-commerce, banking, health care and insurance.

In this article, we focus on an industrial sector where AI is showing great promise: transportation system management. Artificial intelligence can provide traffic system managers with real-time, predictive insights about traffic flow that can lessen congestion and improve safety. However, as explained here, a rugged hardware network must be developed and a software system in place to connect devices and transmit data.

AI in transportation management

Source: IEEE — Smart Town Traffic Management System Using LoRa and Machine Learning Mechanism

Artificial intelligence and machine learning (ML) techniques have gained traction in transportation system management as a key part of Smart City and Intelligent Transportation Systems (ITS) projects and initiatives. Some of the most significant and recent advancements in AI-based transportation networks happened with the integration of sophisticated ML algorithms into ITS technologies. ML algorithms “learn” from studying traffic patterns, pedestrian behaviors and other experiences so they are constantly improving their models and by extension, improving the safety of roadways.

Traffic management is highly complex. The only possibility of dealing with its tsunami of data is to abandon traditional traffic management approaches and turn control over to AI. Artificial intelligence will analyze, summarize and finally relay back to an administrator actionable insight that can reduce congestion, and possibly save lives. Sequenced follow-up actions can range from dispatching emergency services to adjusting traffic signal timing system controls. In most cases, responses are automated without human interaction.

AI’s use in ITS has been heralded as a new era of mobility, one characterized by improved driver safety and comfort, reduced traffic congestion, lowered carbon emissions, and greater speed and efficiency in supply chain management tasks. That’s not to say AI is without challenges. Infrastructure expense, managing competing priorities among city leaders and coordinating the multiple parties and technologies involved in projects have slowed adoption, as have public concerns over privacy and security related to sensing. Despite these challenges, the future of AI in transportation holds immense potential for startups and private companies to pitch their solutions to authorities.

Practical uses of AI in transportation
Artificial intelligence integrated with ITS creates a context-aware solution that merges real-time data from connected road infrastructure with predictive analytics to effectively coordinate traffic across key city arteries. Today, we are on the cusp of a proliferation of traffic applications, technologies and services where AI integration takes center stage, namely:

Toll plazas contain different types of communication, control and monitoring devices to manage the infrastructure effectively. Devices such as vehicle messaging signs, loop detectors, video monitoring, weather stations and toll collection booths provide a great deal of data in a variety of communication mediums that require connectivity.

In all these instances, artificial intelligence gives meaning to data. Better informing decisions with this data means problems can be solved instantaneously by such actions as altering bus and subway schedules, dynamically using weather conditions, adjusting roadway speed limits and traffic light timing, re-routing emergency response vehicles, or applying smart pricing systems on highways, bridges and HOV lanes. Perhaps the simplest example would be video cameras detecting an increase in cars nearing a busy intersection. Automatically extending the green light at the intersection will prevent a traffic jam by dissipating the number of cars at the location of the stop. Still another example would be highway speed limits being lowered in response to sensors detecting a dangerous presence of snow and ice on roads.

Hardware requirements for AI-based transportation systems
When we think of the capabilities and speed of future AI hardware systems, next generation GPUs systems are likely to come to mind. These GPUs perform highly parallel operations and simultaneous computations at breakneck speed when used as accelerators.

Connecting road infrastructure takes much more than GPUs. Gathering instantaneous traffic and lane-related data requires layers of IoT devices capable of autonomously monitoring and broadcasting critical lane and roadway data. Devices include an assortment of different types of sensors, GPS, inductive loops, RFID, pass readers, radar detectors and IP surveillance video cameras, along with integration with connected traffic and systems, lights and smart toll gates.

Below is a partial list of different types of data collected from these devices:
• Real-time traffic conditions
• Car counting
• Accident detection
• Weather and traffic conditions including air quality
• Public events scheduling
• Congestion at traffic light intersections
• Toll booth monitoring
• Ticket cameras that flag unsafe drivers
• Measuring distance between cars
• Presence of bicycles and pedestrians at crosswalks
• Automated license plate reading
• Outside news feeds, social network media mentions and connected vehicles

Dispatching the collected network data from point A to point B is achieved by high-speed industrial switches, wireless routers, power over ethernet (PoE) injectors, media converters, wireless access points and gateways. All this industrial networking equipment is hidden away in roadside cabinets and on top towers. Whether network data is transmitted wirelessly, by cellular, or via ethernet cables, the network data must make its way to a traffic center. That takes hardware.

A roadmap for AI-based its hardware
Behind the scenes making all this automation possible are industrial hardware devices seamlessly performing their duties.

Designing a reliable AI-powered intelligent transportation system starts with hardened roadway networking equipment. Virtually every roadside device is subjected to challenges that can knock connections offline. For this reason, each roadway networking device—from the edge to the control center—must be environmentally hardened, have a reliable power source, and feature robust connectivity and redundancy to transmit data 24/7/365 without interruption.

An “industrial-grade” network device is hardened to comply with international performance standards, such as NECA, ASTM, IEC, UL, CE, ISO, IEEE and EIA/TIA. Commercial-grade components are avoided since these parts are not designed to handle wide temperature swings, heavy vibration, clogging dust, and power surges. This networking equipment is protected by sturdy metal housings rated IP-30 or higher.

Despite their strength, industrial-grade devices for ITS must also remain compact enough to easily fit inside the heavy-duty NEMA enclosures attached to traffic signal poles or concrete slabs. So, add compact dimensions to your list of requirements along with DIN rail mounting, DC voltage inputs, high EFT/ESD protection capabilities, control capabilities, and extended low and high temperature tolerances.

Another key attribute to a successful AI-powered intelligent transportation system is speed. Switches supporting 10G data transmission are necessary for bandwidth heavy collection, transmission, and application of real-time traffic data, therefore reducing the chance of delays or packet loss. Plus, if you are planning on installing a mix of 10G copper and fiber, many 10G industrial switches now feature 1G/10G dual rate SFP+ fiber ports. ITS infrastructures are rapidly displacing copper and coax with fiber for both data and video transmission, so this is a feature that you’ll want in your industrial ethernet switches.

As for redundancy, an open protocol approach is best. Putting interoperation ahead of proprietary protocols means products are easily interoperable with those of other manufacturers. Redundancy protocols are often proprietary. While this may not pose a problem in small networks, in traffic applications that require frequent expansion, proprietary protocols are both time-consuming and expensive. It compels the network administrator to keep building out using proprietary solutions from a single manufacturer, even if those solutions no longer meet their changing needs. Open protocols help you deliver continuous uptime without limiting you to inflexible proprietary solutions. By choosing best-of-breed open-source solutions, you are ensured that no matter where, how, and when you connect, you’ve always got options to ensure your data layer delivers.

What about security? Unfortunately, transportation data is susceptible to cyberattacks that can expose private information or let hackers alter traffic operations with disastrous consequences. Authentication, authorization, and accounting mechanisms will give you the visibility to track activities made by the ITS team and enforce user policies, while limiting security privileges only to authorized personnel. Access Control Lists (ACLs) can further security by restricting unauthorized users from accessing data, systems and devices.

Finally, all those devices mounted in hard-to-reach places like inside pavement, on weather stations, on top bridges, inside tunnels and on other outdoor infrastructure need power. Simply running a wire to the nearest electrical outlet isn’t the answer. That’s why today’s 10/100/1000TX industrial switches offer high-port count PoE technology. Any IEEE 802.3af, 802.3at or 802.3bt (90W) compliant power devices can be safely powered without the need for any additional wiring.

Privacy issues
AI’s skeptics fret that if we run the transportation playbook forward, there is the real possibility that data safety and privacy will be compromised. To prevent this from occurring, it is important that each development in AI offers steps forward in transparency, stronger governance, adherence to applicable laws and regulatory compliance to manage potential risks in bias, discrimination and privacy violations. An ecosystem of tools around safety, compliance and privacy is emerging as AI gains further traction in transportation applications. Finding balance will require new anti-privacy rules to be constructed in a way that allows innovation to flourish and supports a level playing field.

Henry Martel, field application engineer, Antaira, has more than 10 years of IT experience along with skills in system administration, network administration, telecommunications, and infrastructure management. He has also been a part of management teams that oversaw the installation of new technologies on public works projects, hospitals, and major retail chains.