The City of Helsinki is piloting solutions for more real-time traffic planning and control and open data provision
The City of Helsinki wants to improve its capability of obtaining a real-time view of traffic from multiple data sources. A real-time view of traffic is a graphic view of data related to traffic, lanes, the locations of means of transport, and conditions. It can provide a short-term forecast of traffic volumes, the smoothness of traffic, incidents, punctuality and prevailing conditions.
Tieto built the system required for the pilot project, supplied the equipment and developed a video analysis solution in collaboration with Tampere University of Technology.
Why is a real-time view of traffic needed?
Two main purposes were identified for traffic video analysis:
Analysis based on history data to support planning and decision making (data is collected in non-real-time over a longer period)
Analysis of real-time data to obtain an accurate view of incidents locally
Situation data based on a real-time view plays a central role in improving the smoothness and predictability of daily journeys and transportation as well as traffic safety. Data can be transmitted to end users, service providers and others who can benefit from it.
Traffic data collected over a longer period can be analysed so as to draw up forecasts of traffic flows and the effects of various incidents compared with the normal situation. Examples of incidents are traffic jams caused by major public events and accidents as well as poor weather conditions. Forecasts help in guiding traffic effectively immediately after an incident and preparing for the consequences.
Real-time analysis of traffic data also enables better services for city residents. Data on vacant parking spaces can be distributed to motorists involved in the system. Video material also reveals when it is time to clear snow from bicycle and pedestrian paths. The Finnish Transport Agency collects travel time data, requiring a lag of no more than one minute from the actual situation.
The goal: cost-effective data collection and analysis
Since a large number of mobile video recording units are needed to obtain a reliable and comprehensive real-time view, the system must be designed using as inexpensive equipment as possible.
That is why low-cost Android phones were selected as cameras that were attached to vehicle windscreens. Mobile phones are equipped with cameras of sufficiently high quality and built-in GPS tracking, making it easy to synchronize images with locations.
Phone sensors also provide data on acceleration, deceleration, vibration and other environmental factors that influence the usability of data. Off-the-shelf software is available and software customization is easy.
It is expected that the investment costs needed to improve the quality of the real-time view can be recovered through reduced costs of accidents, congestion and emissions, as well as through new business operations. The real-time view makes it possible to address incidents due to their rapid identification, positioning and forecasting.
Lessons learned form 60 000 observations
Some 60 000 observations were obtained for assessing the models. The system identified and classified the passenger cars seen in the videos with an accuracy of up to 95 percent. The identification percentages of other objects were lower.
To be useful in practice, the video analysis models need to be further developed. The analytics can be trained for desired purposes in order to increase the accuracy of results and expand the solution’s potential uses.
Today’s technology is not yet mature enough for real-time video analysis of big data. The main reasons for this are data transfer challenges (4G networks have insufficient capacity, so 5G networks are a must) and large data processing needs. If the real-time requirements are sacrificed, the solution is feasible as is.
Data transfer would not be a problem if analysis could be performed by the recording device. In that case, it would not be necessary to transfer all video material over the network for analysis; it would be enough to transfer metadata from local analysis, which requires much less capacity.
The future looks bright with data
The trial was unique in that data produced by mobile cameras has not been utilized very much to date. There were some flaws during the pilot phase – which was to be expected – but the results were interesting and quite encouraging, considering practical applications. The project provided valuable information on how methods of video image interpretation work and how suitable existing video recording systems are for producing traffic data.
From the perspective of further development, the availability, quality and analysis of data can be improved in several ways.
With sufficient incentives, it may be possible to crowdsource data collection – i.e. involve motorists in collecting data for analysis – by offering cameras with video analysis capabilities for private cars.
The number of static cameras at critical spots can be increased to help collect useful data.
Cameras can be installed on public transport to generate traffic data on the entire route network.
External data sources, such as accident data from emergency response centres, can be connected to the system to improve predictability and accuracy.
As the data-processing capacity of smartphones increases and the 5G data transfer system evolves, it will become possible to spread analytics over the entire ecosystem and, thereby, better meet the real-time requirements.
The City of Helsinki and Tieto will continue to look into new technologies and opportunities to leverage data, with the aim of providing better services for city residents.