DA-SPACE - Road Traffic Monitoring Solution for Smart Cities

25-05-2019

Team Traffic Watch consists of eight students from Faculty of Informatics and Information Technologies (FIIT), Slovak University of Technology (STU) in Bratislava that developed the winning solution within the second cycle of the Interreg project DA-SPACE. Their goal was to create a system that provides knowledge for variety of road-traffic management & planning decisions. The solution was developed in cooperation with companies Unicorn Systems SK and Orange SK that offered the challenge to the team and provided the mentoring support to the team together with Dr. Ivan Srba from the FIIT. Professor Marian Zajko (STU Institute of Management) talked to three members of the team Adam Kňaze, Jozef Kamenický and Matej Horváth before their presentation of this solution at the Innovation conference taking place within the Interreg project DA-SPACE on 15 May 2019 in Ulm, Germany.

Team Traffic Watch - winner in the 2nd DA-SPACE Local Demo Day in Bratislava

Q: Why do we need to monitor traffic?

A: Here are some situations when it would be needed:

  • Heavily used street in the city centre is going to be closed due to construction works. An alternate route has to be found to prevent a traffic collapse. Prerequisities are: analyses of how many vehicles actually need to be diverted, what are the capacities of possible detours and what are their current traffic levels.
  • One of the intersections in the city is congested. It may be rebuilt and expanded. Firstly, the cause of the congestion needs to be identified. Therefore, it is important to know what are currently the most heavily used areas of the intersection and where do cars spend the most time waiting.
  • Routes and timetables for public transportation are created in order to reduce car usage. The starting point is the knowledge of how many cars are transiting to particular borough at what time of the day.

Q: How does your solution work?

A: Our solution consists of multiple parts. First part is the network of smart cameras. As does the word “smart” indicate, these cameras do much more than just record a video. Using computer vision algorithms, they detect passing vehicles and analyse their movement.. The device we are using for this is a Nvidia Jetson TX2, one of the few edge devices that support graphic card acceleration.

Second part of our solution is the application server connected to a database. Application server receives information from cameras and stores them in database. It also creates hourly statistics. More difficult tasks involving filtering and aggregating by time are executed very quickly and efficiently thanks to time-series database TimescaleDB.

Third part is a web application that acts both as a admin interface for cameras and as a user interface to present gathered data in an insightful and user-friendly way.

Presentation of theTraffic Watch solution in the Innovation Conference in Ulm

Q: What about existing solutions?

A: Existing solutions differ in the level of technology involved, their accuracy and installation cost. Basic human-based counting with pencil and paper (which is still the most common solution in our region) tends to be inaccurate and not really suitable for long term monitoring. Mechanical devices physically installed on the road disturb traffic during installation and are not easily re-installed to different locations. There are computer-vision-based systems, too. Another group of products does trajectory analysis and speed detection. However, they usually use expensive equipment with radar.

Q: What are the strengths of your solution?

A: Our solution has several advantages. It is easily deployable and maintainable via remote web admin interface. Use of edge-computing ensures good scalability. When it comes to delivered insights, we are providing a package of multiple different analyses based on one shared AI core. Last but not least, all of this is computed and displayed in real time.

Q: So what are these analyses you keep mentioning?

A: First group of analyses uses “transits”. Users may define several zones in the camera view. When a vehicle passes the view, we track its movement and log all the zones it passed and times of passage. For example on an intersection, user may mark all points of entry and exit. Our system then provides him not only with information about how many vehicles at what time entered the intersection from what direction, but also where they were heading.

This gives him a general sense of traffic flow and its trends.

We are working on further extension of transit information in several useful ways. For example, for each vehicle a type classification will be performed (to the predefined vehicle types, e.g. car, van, bus, truck). We also record an average time of passage between zones. This is useful because pure traffic volume itself does not help in identifying the traffic congestion. If traffic is congested, number of passing vehicles is low (they move slowly), which could be falsely interpreted just as a low amount of vehicles with no congestion.

Transit events mentioned until now ignore precise trajectories of vehicles while moving between zones. However, we do record and evaluate these as well. It gives us more information about the quality of design of observed road. We record and display:

  • heatmap showing volume of passing vehicles
  • precise trajectories of passing vehicles
  • heatmap showing stationary traffic (vehicles stopped/waiting)
  • heatmap showing relative speeds of vehicles

This kind of information shows general driving behavior of passing vehicles and may possibly identify problematic or dangerous areas.

Q: Does it work in real life?

A: Yes, we used public web-cam stream that observed crossroad located in Podolsk, Russia. We have been monitoring it for one week. Apart from test of long-term stability of our system, we analyzed traffic and came up with several observations: vehicles driving to near residential area were often unnecessary halted by another passing vehicle, which may be solved by separate lane to residential area. We also found out that drivers ignoring traffic rules are causing delays, which can be solved by additional traffic lights.


Q: What are your future plans with the project?

A: In the future we plan to take it one step further and perform vehicle re-identification. It means multiple cameras on different locations with ability to identify if a particular car had previously appeared on another camera. We plan to do this without reading vehicle registration plates because it requires expensive or specially positioned cameras. Our solution would be the first and only one with this functionality.

Thank you  for clarification of your contribution to building Smart city in Bratislava and I wish you all the success in the Innovation conference in Ulm.

Programme co-funded by European Union funds (ERDF, IPA, ENI)