VIREBOX - case Škoda
Tactful passenger counting in Tampere
If the passenger counting system developed for the Tampere trams learns what it is expected to learn, Kai Hermonen, who is responsible for Škoda Transtech’s information management and SmartRail ecosystem project, sees huge potential for its use.
Škoda Transtech Oy manufactures rail equipment for demanding conditions. It specializes in the production of double-decker passenger cars and low-floor trams. Škoda Transtech heard about Vire Labs in search of a partner capable of analysing and interpreting video image of their ecosystem.
“Our experience of the passenger counting systems used in the past was far too rigid. The solutions required the installation of separate equipment in every doorway, which posed challenges in terms of design, installation, and cost. The problem with the solution was that our customers were unable to use counting on all trams. We began to wonder whether passenger counting could use existing surveillance cameras, as there were already 12-15 of those on each tram,” Hermonen describes as the starting point.
Vire had the will and the guts to start developing something globally unique
Kai Hermonen, Škoda Transtech
The development of machine vision is a collaboration
For the SmartRail project supported by Business Finland, Transtech mapped out the kinds of companies utilizing machine vision that can be found on the Finnish market. Hermonen divides the actors into three groups. There are companies that have skills, but are not prepared to invest in the innovation, research, and development phase of their own work if the finished payer is not yet confirmed. Secondly, there are research institutes which have a lot of research and development activities in different machine-vision solutions, but not a solution directly suited to this need.
According to Hermonen, the third group includes teams like Vire Labs who do not necessarily have a finished product but have even more experience with radar or other smart-based solutions and the desire to jump towards new things. These operators often also have knowhow of “iron”, which crushes data. Transtech ended up at Vire Labs because Vire had the will and the guts to start developing something globally unique with them.
As the development progresses, new requirements from the customer, the test environment, and the platform are constantly coming that need to be somehow met. “Our collaboration is such that Vire Labs solves 70-80% of the problems. We bring 90% of the problems,” Hermonen jokes. “It has gone well. Research and testing have been carried out on both sides. We know at both ends that this product is not yet ready, but the vision of where we are going is the same. The aim is to achieve a collaboration that both will benefit from,” says Hermonen.
The world's first real-time passenger information system
“What we are building does not yet exist anywhere in the world. It’s about real-time passenger information. When leaving a stop, we are immediately informed of how many passengers are in each part of the tram. We can categorize the passengers with the accuracy of human, dog, bike, and stroller. In addition, we get a ‘stop flow’ in almost real time for all trams, which means we can see how the entire tram traffic goes,” says Hermonen. The calculation of passenger capacity is carried out with edge computing, in this case on the tram. Video image never leaves the tram. Vire Labs Virebox.AI combines sensor information collected from the tram with the data from the camera and initiates edge computing. As a result of the computing, only the number of passengers and identified objects in each car is transmitted to the cloud for analysis. “If a system that meets GDPR regulations is built in Finland, it can also be easily deployed on the international market,” says Hermonen.
...it can also be easily deployed on the international market
Kai Hermonen, Škoda Transtech
Neural networks need real conditions to learn
Another thing Transtech customers want is a solution to identify disruptive behaviour in the video image. This is a typical example of a challenge that Transtech throws forward to Vire Labs.
The biggest part of the work is coming ahead in the autumn, when the new system is set up for normal tram services in Tampere. The rehearsal has been done, but only under real conditions can the system learn what it’s like to recognize figures and objects at different times of day and year.
“When the system is operational in Tampere, we will also look at a possible test site from the Helsinki metropolitan area’s rail traffic and we will build a platform there that will benefit all Škoda Group customers. So, exciting things are happening,” sums up Hermonen.