Curbing Noise Pollution with AI-Enabled Solution

(Inaudible) City noises can be very annoying, but they're not just annoying. Noise pollution poses an environmental and health threat. And as cities continue to grow and urbanize, the issue only increases. For example, in New York City, noise complaints went up by 83% from June 2019 to June 2020. Many cities have taken measures. The city of Stamford, for example, became more punitive regarding the issue in 2023, increasing fine prices from $90 an incident to $250 an hour for bars and restaurants exceeding sound levels. And Sant Cugat, near Barcelona, deployed a solution that made lights flicker when the volume around them was too loud. Most cities experience noise pollution at some level from traffic, social events, construction, industry... And Newport, Rhode Island, is not immune to the problem. The 25,000 resident city gets about 4 million visitors a year, which can cause a lot of congestion and therefore, a lot of noise, leading the municipality to look for a smart city solution to curb the issue. We started out where we were hearing a lot of complaints from the local populace, especially in the summertime, about vehicle noise, specifically motorcycles. And the more we started to look at that problem and how difficult it was to control it and even enforce it, we tried to, we found, for example, that enforcing it was very difficult and very inefficient. So we would take a police officer and stand them on a street corner with a decibel meter, which was a very inefficient use of that resource, so we were looking for something that was easier to control. The city administration, in partnership with the police, turned to a smart solution by Sorama. They deployed what are dubbed as noise cameras in problematic areas, and started with a proof of concept, gathering data that they reviewed every week with the police to better understand what the issues were when it came to noise. The basis is a microphone array a MEMS microphone array. So typically several tens, maybe sometimes even hundreds of microphones in a grid combined with a video camera processing power on the edge, as we call that, on the device itself to translate the sounds that those microphones are picking up, into an overlay on the video screen where that sound is coming from. So it's a bit comparable to a thermal imaging camera that you might know that shows temperature differences. It’s location, what is it and how bad is it? For the project, the acoustic sensors, which work thanks to power over ethernet, were connected to AXIS license plate readers. The city deployed one unit on a pole, so it’s stationary, and then deployed two others on mobile trailers, something the municipality built itself as they realized they needed to be able to change locations often. When the solution detects a sound that exceeds the volume limit, the camera is activated and takes video shots of the vehicle, a process that is possible in some respects thanks to AI in the form of visionAI in the cameras and machine learning to classify the noises. Embedded in the acoustic camera we have GPUs, running with AI course on it. And one of the most basic ones is an untrained anomaly detector. Unsupervised, the model can train on the device itself, and you simply say, okay, let's leave it on for a minute, an hour a day. all the sounds that are being captured are considered normal sounds. Then we automatically trigger or we trigger a device to go into detection mode, and then any significant anomalies to the sound it has been trained for will trigger an alert. We also have pre learned models. When an event is detected, the data details, which include video, audio, Sorama sound map, decibel levels, license plate and vehicle type among others, are sent to the AXIS camera station and stored in the Data Factory Platform by FF Group. A police department operator, reviews it and decides whether to fine or just send a warning. The platform is to serve as a centralized data management platform for what, in the future, is set to become their smart city strategy. During the PoC, the municipality only used the solution to gather data and warn. But now the project is set to go live regarding enforcement as of July 1st this year, which means tickets. Starting at $250 for first time offender vehicles, fines will be issued to those that have been altered to make them noisier. That was a point that, in our conversation, Jim wanted to highlight. Tickets will target specific rule breakers, not all of them. All the other solutions that we looked at, the few that are available, represent a business model that we didn’t really like, which was, without question, if you surpass the audits, you're going to get a fine. It's going to be sent out by some third party. So what we really wanted was, I guess, the extra cost of having an officer review every occurrence. We're not going after the resident who's driving back and forth to work with a broken muffler. What we’re going after are the people who are intentionally modifying their devices, on purpose, to make them very loud. We're going after vehicles that are blaring music at the wrong time of day, like night time, way too loud that sort of thing. As the PoC proved successful, the city is set to increase the number of noise cameras to between 5 and 7 units during 2024. The project has already led the city to change its ordinance. Before the noise detection solution, the decibel level for vehicles was 86dB, but after analyzing the data gathered, Newport has brought that number down to 83. Although many vehicles and situations will be exempt from the fines, and enforcement is not the main goal of the project, issuing fines will encourage people to behave, but also, will allow the city to have a return on investment. It was through the city administration that we used project funds, which are generally available for technology endeavors like this. So we used technology funds 100% so far, actually, until we go live. And then it will be dispersed on the regular police department budgets for as far as maintenance of the trailers, that sort of thing. This may look like a very niche smart city project, as it's quite small and specific, but it's the beginning of Newport's smart city adventure. It's a niche solution, coupling computer vision with the cameras and machine learning with the noise detection, two AI uses, and as we know, AI is becoming key in the technology world and in the future of smart cities. The Rhode Island city was looking to start a strategy, which is why they decided to begin with a smaller use case with a high profile technology to learn and develop a centralized smart city data platform through solving a city issue. They want the platform at the moment intended for their noise units to act as a centralized data management system. So if in the future they deploy other sensors like air quality sensors, they can pair it to and analyze it with the other data gathered. This was their first step in the smart city arena, which is why they started small, slowly but steady, helping city managers envision the possibilities of having a city that is smart.