Tunnel operators benefit from AI-powered automatic incident detection
13/06/2023
Camera-based incident detection has become an established technology to support tunnel operators in keeping their roads safe and in organising a fast response. After the introduction of thermal cameras as a solution for 24/7 automatic incident detection, systems supported by artificial intelligence (AI) may well be the next wave. Today, AI is already bringing incident detection to an even higher performance level. 
In February 2023, the European Commission published preliminary figures on road fatalities for 2022. Around 20,600 people were killed in road crashes last year, a 3% increase on 2021 as traffic levels recovered after the pandemic. Although this number is 10% below that of 2019, one year before the pandemic, the EU still has an important target to achieve, as it intends to halve the number of road deaths by 2030.
Research shows that there are typically fewer incidents in tunnels compared to open roads, but it is also known that the severity of such incidents is higher. In 2004, a European directive on road tunnel safety prompted the roll-out of incident detection systems in tunnels and since then the technology evolution has not stopped. With the ongoing advancement of video analytics-based systems, tunnels have become safer. In case of an incident, first-response teams can now be deployed minutes, or even seconds, after an incident or an irregularity (such as a fallen object, a pedestrian or a car slowing down) has been detected.
Thermal cameras have become an established technology and even a critical asset for operators to guarantee accurate detection throughout the tunnel infrastructure. The use of thermal imaging cameras has proven particularly valuable for tunnel entrances and exits. In these locations, shadows or direct sunlight can obstruct the view of visible-light cameras and therefore disturb traffic detection. As they detect heat, not light, thermal cameras have no issues with these phenomena and, as a result, can detect traffic 24/7 in all weather conditions.
One of the biggest advantages of thermal imaging cameras in the field of tunnel safety is that they can effectively see through many types of smoke. This makes them an ideal technology for tunnel safety operators or emergency response teams to find their way through smoke-filled tunnels or for incident detection systems to spot incidents in time.
Thermal cameras are also unaffected by headlights, which conventional closed-circuit television (CCTV) systems typically have difficulty with. Headlights can generate false or missed calls and make accurate observation of highway traffic at night impossible. At night, a road can look like an indistinct row of lights to a video camera, making meaningful data collection and incident assessment impossible. However, thermal cameras see the heat signatures of vehicles clearly from miles away. Thermal cameras are also an ideal solution for the detection of fires in tunnels at an early stage, even within seconds of the appearance of visible flames. This allows traffic operators to immediately close the tunnel and take action to quickly extinguish the fire.
Both visual and thermal cameras have their merits. A visual camera may provide operators with more detail to assess the nature of an incident, while thermal cameras have proven to be unbeatable in detecting incidents in more challenging weather conditions. Currently, both detection technologies are often combined into one system, thereby offering operators the best of both worlds. Teledyne FLIR’s ITS-Series Dual AID camera is an example of such a system.
It is impossible to ignore the huge contribution AI technology is making to incident detection. Over recent decades, computing speeds have increased, prices have dropped and the exponential growth of data has worked to improve AI and make it more efficient. Instead of conventional rule-based detection, data-based systems are now leading the way. These systems can be trained on large datasets of images and learn how to identify and classify objects in an image. They use this knowledge to make decisions based on new images that they have never encountered before.
Previous generations of traffic detectors have always looked at subtle colour differences in the image at the pixel level whereas the AI-powered detectors of today are looking at the entire camera image to make ‘intelligent’ decisions. In short, AI systems can handle more complex traffic situations and are much better at making smart predictions.
AI-powered systems outperform non-AI detectors in a number of ways. First, the accuracy of detection is much higher with AI detectors. Though good detection rates have already been achieved using conventional systems, AI-powered detectors can take away even more of the frustration caused by unwanted alarms, meaning response teams are not sent out erroneously.
AI-based systems are more successful in detecting different vehicle classes. Detectors from Teledyne FLIR can easily distinguish between a car and a van, or between a small truck and a large truck, and can even be taught to recognise customised classes. The system only needs to be fed with new data and trained. AI-based systems can also more easily distinguish between a fallen object and materials used in roadworks. With cameras so smart, installers can be more flexible in installing their equipment. The position of camera units is no longer limited to a certain height. Even in less ideal camera positions, the detection performance of AI-based systems remains high.
AI detectors are also better at predicting trajectories. Based on vehicle parameters such as speed and direction, they can easily determine where a car is going, even if for part of that trajectory the view of the car is occluded by a passing truck. This makes detection much faster and more accurate. Operators can be warned by so-called pre-alarms for cars that are slowing down and likely to cause collisions. Traffic detectors from Teledyne FLIR can also keep an alarm active until the situation is cleared, for example until a stopped car has left, a fallen object has been removed or a pedestrian has exited.
Another promising application of AI-based systems is that they enable a digital twin of the tunnel environment to be set up. The digital twin is a concept that has already been popular in manufacturing and production industries for many years, but instead of the virtual real-time representation of business processes, tunnel operators can use the digital twin to obtain a complete overview of all traffic operations and activities in a tunnel. A tunnel is a complex collection and collaboration of different systems and technologies, including safety management, ventilation, video management systems (VMSs), communications and many more. A digital twin establishes the exchange of data between these physical tunnel systems and their virtual representation. Based on this wealth of data, a digital twin can generate a real-time bird’s eye view of the traffic inside the tunnel, offering operators an invaluable source for decision-making.
Data-based detection systems very quickly become the norm, so it is easy to understand that a system’s detection performance will soon be determined by the quality of the data it is trained with. High-performance systems will need a lot of data for training, in this case video images of traffic. Teledyne FLIR is a company with a heritage of 30 years of in-house data collection and believes that data is the true quality mark for high-performance accurate traffic detection.
AI-based camera products such as Teledyne FLIR’s TrafiBot series also have their AI embedded in the camera. This means that it is not 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. AI solutions from Teledyne FLIR have redundant connections to the tunnel’s programmable logic controller (PLC) system, which means they will continue to detect even when the network is down.
Today sees the beginning of a new wave of AI-based systems and there is no doubt that detection performance will improve even more in the future, especially in the difficult corner cases where conventional systems often reach their limits. With possibilities for traffic operators to join forces by connecting their tunnel and highway data with sensors from city traffic, an even more complete and robust image of traffic may be obtained. In any case, there is a great future ahead for AI in intelligent transportation monitoring systems.
