Peter Mildon, COO and Co-Founder of Vivacity Labs, has been reviewing data from their national network of AI based video road sensors to assess the impact of Covid-19 on our Highways Networks on a daily basis.
In this article Peter provides:
- A brief introduction to Vivacity’s technology for context
- A dive into the impact that Covid-19 has had on transport habits across a range of road-based modes and geographic locations
- Details on how Vivacity have adapted their sensors to measure social distancing and the results
- Conclusions for the Post-Covid-19 world
Introduction to Vivacity’s Technology
Vivacity provide permanent smart video sensors that offer fully classified road usage data throughout the year. The principle advantage of using video as our sensing technology over any other methodology is the richness in classifications possible to obtain. No induction loop or magnetometer can count pedestrians, and no radar or thermal camera can classify the difference between a car or a taxi! When working with the richness of video data, all these things become possible (Figure 1).
Figure 1: Vivacity’s Machine Learning detector providing classification of pedestrians, cyclists, cars, motorbikes, buses and taxis within this example field of view.
A second key advantage of using a video-based sensor is that data can be gathered across a wider area of road, typically 30m square – meaning the path (Figure 2), route and behaviour of pedestrians and vehicles can be determined, as well as enabling multiple flow count lines to be placed in the field of view of a single sensor.
Figure 2: Path output from a Vivacity Sensor, showing the path of cars (red) and pedestrians (purple) at a complex junction.
The benefits of this system have already been clearly seen by large parts of the market, with our client base growing to over 15 cities – with data being used for a variety of use cases from transport planning through to real-time network management.
Traffic Volume Changes since 16th March
We have been providing the DfT with an analysis of our data every couple of days since the Lockdown started.
In this article, I have drawn out some of the conclusions for a wider audience based on data from 10 cities and one motorway, covering a total of 362 count lines. This data set is a sample from our most mature and established sensors, and where our customers have consented to the data being shared.
We took the average traffic flows for each day of the week for the 6 weeks from 3rd February to provide a pre-lockdown baseline. All data is presented as a change against this baseline.
Consistent with other traffic measurement methods, we have seen a 60-70% reduction flow since 9th March:
Figure 4: National traffic flow since 9th March 2020
Splitting this analysis out by geographic region has shown a consistent view from city to city:
Figure 5: Change in traffic flow by city
If this data is split out by class, however, interesting differences become apparent:
Figure 6: Change in traffic flow by mode
- Pedestrian levels have seen the biggest decrease, with the exception of weekends, where car traffic has reduced marginally more, while pedestrian traffic tends to spike slightly
- Cyclist numbers are down 60% on week days, but over the weekend of 4th-5th April and 11th-12th April, they were actually higher than during the baseline period. This can be attributed to the good weather, and the number of people taking their daily exercise on bikes. It is worth noting further that our data sample is slightly skewed towards Cambridge and Oxford (Figure 3), where cycling levels were very high during the baseline due to the student population, but have since reduced significantly as the students have been sent home. If Cambridge and Oxford are excluded from this analysis, the weekend peaks in cycling volumes hit between 2x and 3x of the pre-lockdown levels.
- Commercial vehicle traffic is down between 30% (OGV1 – Articulated HGVs) and 40% (OGV2 – Rigid HGVs and LGVs). This is consistent with the CBI’s estimate that the UK economy shrunk by 35% compared with the period immediately prior to the Lockdown.
- Only 40% of the pre-Lockdown bus volumes are still running. The reduction in buses was seen in a weekly step each Monday, rather than the gradual decline seen by other classes. This is consistent with bus companies planning their timetables on a weekly basis.
Direct Measurement of Social Distancing
In Oxfordshire, Bournemouth, Manchester and Peterborough, we decided to push the analysis further. Our sensor network here has not just been recording the volume of traffic, but also the path that each road user was taking across the space. We decided to post-process the data to calculate if social distancing measures were being followed.
Figure 7: Example of measurement between pedestrians calculated with pedestrians coming within 2m of each other counted from stock footage. Red shows <2m; Yellow 2-3m; and green >3m
It has been discussed a lot in the media recently whether the government should be using mobile data to monitor social distancing. Many people are concerned that such an invasion of civil liberties might not be un-done once the crisis was over. This type of video analysis provides an alternative, which is non-invasive from a privacy perspective and offers a much higher resolution on the social distancing measurement. By using edge processing, no personal data is ever generated by the system – no videos are transmitted or stored, and the AI never knows who it saw. Instead we simply are able to acquire the data that is needed to help the Government make decisions on how its lockdown policy should adapt next.
The analysis across 4 cities showed a much bigger variation in the change in number of social interactions by location than was experienced when traffic flow as a whole was considered (Figure 5 showing geographic similarity compared with 8 showing variation). We hypothesised that this was being driven by the type of road where the sensors were located, rather than differences between the populations of the city. While a typical vehicle journey will pass through many different types of road, the same is not true of a short walk, which will be concentrated closer to residential areas. Furthermore, features such as the width of the pavement will have significant impact on the number of <2m interactions. It is easier for pedestrians to keep a distance in a pedestrianised city centre, compared to on a narrow pavement alongside a trunk road.
Figure 8: Change in <2m social interactions by city
We therefore split our data out, not by city, but by proximity to the city centre, and found some interesting conclusions:
Figure 9: Change in <2m social interactions by urban area
- Social interactions in Oxford city centre have reduced by 90%. This is consistent with the drop seen in Manchester, where our sensors are also located in the city centre. The closure of shops, restaurants, bars and clubs, along with the ban on non-essential travel is clearly keeping people away from these areas.
- Residential areas and local centres have seen a drop in interactions of about 70%. The higher residual number can be explained by the necessary trips people take from residential areas to local centres in order to buy food and take exercise.
Figure 10: Change in <2m interactions by commercial land use
- The number of <2m interactions at retail parks is consistent with those seen in residential areas and local centres.
- The closure of university campus areas has led to a drop in interactions of 90%
- Business parks show a completely different trend:
- Weekday reductions are only down 60%, indicating that a proportion of the workforce is still coming to these locations, and are not able to maintain the <2m rule when they do
- Weekend spikes in these areas do not indicate a high volume of interactions, as the baseline was already very low. Instead it would appear that these historic weekend interactions were in someway “essential” as they have only reduced by 5%.
Conclusions for the Post-Covid-19 world
The analysis we have been carrying out has highlighted the benefit of permanent data collection over temporary counts. The industry has typically built transport strategy models based on surveys carried out on “normal” days, deliberately avoiding Mondays and Fridays, or bank holidays and school holidays. While this approach helps reduce the amount of data that needs collecting, it also clouds the picture of the true demand for the road network, and leaves it susceptible to significant errors due to normal fluctuations in demand.
In a world where active travel modes are being strongly pushed by many authorities, the variability of pedestrian and cyclist traffic from one week to the next based on the weather should serve as a perfect example of this. These are classes that are also underserved by the more traditional induction loop sensors that have dominated the Automatic Traffic Counter (ATC) market in the past decades.
The rich path data output from the sensors also have uses beyond what even we imagined when we developed it. The extension from looking at desire lines to calculating social distancing data can easily be extended to measure the proximity of HGVs to Cyclists, or the number of tailgating vehicles on a high-speed road.
Under the current Covid-19 lockdown, Vivacity’s system has become invaluable, with our early-adopter clients reaping additional rewards from their investment:
- Our API has been directly integrated with DfT’s data amalgamation efforts, making our client’s reporting responsibilities painless
- We are the only sensor offering pedestrian and cyclist data alongside classified vehicular data
- Our existing sensors have been updated to measure compliance with social distancing measures
- The value of permanent counters against manual surveys is particularly stark in this climate – during the first 2 weeks of the lockdown, transport behaviours changed on a daily basis. As the Lockdown measures are unwound, the same will become true again. This system provides real time feedback on how the public are responding to the Government’s measured – all at a time when temporary traffic surveys have typically been cancelled due to the abnormality in traffic flow conditions.
Finally, as more and more cities adopt this technology across the country (and the World), it enables the type of national analysis presented in this article, something which the DfT previously did not have access to. It also offers each Local Authority the opportunity to benchmark their active travel strategies, or modal shift indicatives with others around the country.
If you would like to discuss the findings in this article, or want to find out more about our technology, please contact me at firstname.lastname@example.org
Details of where Vivacity can provide solutions throughout cities can be found at www.vivacitylabs.com/solutionsmap/
For additional insights on “Revolutionising Data Collection and Traffic Optimisation Using AI”, use this Retrospective to view slides from, CEO and Co-Founder, Mark Nicholson’s presentation at TRANSPORT Smart Class, Wales 2020!