How Fyma turned cameras into sensors for AI-based traffic analytics in Dubai
Here at Fyma we believe that the millions of cameras sitting idly around us, providing at most some emergency ‘after the fact’ analysis or simple statistics capabilities at best, could be harnessed in a secure and safe way to create value from the images the cameras are seeing. There is so much unused yet valuable data out there that is simply not being harnessed to create value. What we are talking about are things like whether a car crash is going to happen or a crime might be committed for example. Fyma is here to turn cameras into predictive sensors and turn these ‘dumb’ devices into smart and helpful ones.
During the beginning of 2020 Fyma (formerly Visory) was selected by the Dubai Future Foundation as one of the companies to participate in a governmental project to provide smart monitoring systems for the Dubai Road and Transport Authority. You can read more about the challenge here:
Some background to the project
During our time in Dubai at the beginning of 2020, the whole world changed because of the COVID-19 outbreak. And precisely because of the COVID-19 outbreak, camera-based automated intelligence became a lot more important in ensuring our safety and wellbeing. This case study is not going to be about the ongoing pandemic, don’t worry. But we do need to touch on this topic as there is no denying that ever since the COVID-19 pandemic swept across the world, the chance of any of us returning back to our former everyday lives anytime soon, if at all, is a big fat zero. When a big world-changing event like this happens there is always a lack of two things: up-to-date data and common sense.
Putting the pandemic aside for a second, having a real-world digital trace of what is happening in any environment is key to making the right choices and creating new businesses and opportunities. It is just that when COVID 19 hit, this became really visible as we had no idea how people were moving and interacting in urban environments. This, however, had a direct impact on how the virus spread and how cities, for example, needed to react. Cities and everyone else for that matter need data. This is something that the AI world deals with every day: we always need more data, cleaner data and we need it as often as possible. This is not always easy or even possible. Think about hardware sensors for example. A sensor that can capture videos, sound, temperature, humidity and much more can cost from 50$ up to more than 4000$. While one might argue that the data is worth putting hundreds of cameras like this on the streets, the counter-argument is that cities and companies simply don’t have the funds to do that, especially now.
So how can we collect data while at the same time not bankrupting everyone? Well, why not reuse existing infrastructure, and by infrastructure I mean cameras. While there are a lot of data sources that can be used such as mobile phone movement data from telecom companies and flight records, one data source is being overlooked: urban video cameras.
These cameras are nothing new and have been used to monitor traffic and crime for more than 10 years, but with the implementation of computer vision AI, it is possible to understand the intent and context of the movement of individuals in a meaningful way.
How Fyma was used in Dubai?
The implementation of Fyma in Dubai had two main goals:
- Provide intelligent monitoring solutions around traffic counting and classification — How much traffic is on the road at any given time and what type of traffic is it (i.e how many cars, pedestrians, busses, etc.)
2. Identify potential safety issues and anomalies.
While the first object seems pretty straightforward for any AI computer vision based system the second one is a lot more complex. Why? Well instead of just counting and classifying objects we need to actually look at the behaviour and intent of all objects that the camera sees. Also, this system needs to learn the normal behaviour for each camera that it sees before it knows what is abnormal behaviour or before it can begin to analyse intent. We started by looking at camera streams and what they are currently used for and then seeing how AI can complement that. The results were intriguing.
We found out that it is actually really easy to take a regular video stream, direct it towards our platform and have it start sending valuable but anonymous information about urban movement. Fyma started learning each specific camera feed that it received and started giving out the basic statistics such as traffic count and classifications of traffic types. On top of that, Fyma uses dynamic trajectory mapping of each object it sees to understand the paths each object travels. This is then used to learn the “normal” movement patterns of each specific camera. This is now actionable data because it can be used to automatically detect anomalies such as illegal U-turns and lane changes as well as other anomalies that may occur.
The outcomes and statistics gathered during the pilot:
- Increase / Decrease of pedestrians on the streets
- Distance between pedestrians
- Amount of car traffic
- Amount of cyclists
- Amount of light goods vehicles
- Amount of buses on the streets.
- Anomalies in the streets
- Standard/non-standard behaviour on the streets.
The outcome for the project — to get real-time statistics and analytics from a large and busy junction for further decision making — was achieved with minimal effort. The surprise for us was the number of all the different government agencies that can use the data, from the municipality who built the road to RTA itself who needs to monitor traffic throughput to the health authority that can monitor whether people are maintaining social distancing rules when interacting in public. This also means that the cost for our system could be spread out between all the entities that benefit from it, making the positive impact felt on a much wider scale while being truly value for money.
Fyma is an emerging leader in the visual information analysis field. The company’s AI platform creates intelligence in cameras, enabling sensor-based analytics and automating visual data from otherwise dumb devices. The company has developed a privacy-by-design approach that anonymizes data and protects sensitive information. The AI does not by choice, for example, enable facial recognition, and delivers intelligent planning based on camera data.