GoodVision Vs Manual Traffic Counters: Comparing User Experience on Traffic Surveys
Published by Daniel Stofan on
Almost all traffic survey companies have previous experience with manual traffic counting. Many manual counters claim they can be super-efficient with it, count traffic manually from videos at 4x-6x, sometimes even 8x-16x speeds, which they claim is more cost-effective. I think this is barely possible to do accurately.
Everyone who has ever seen a video running at 6x or more significant speed must admit that it is barely possible to spot all traffic objects on multiple movements and categorize them into the correct vehicle category. There are many areas where the user experience in manual traffic counting and GoodVision’s automated traffic collection using video analysis differs. So we decided to put these two methods to the test and summarize the results in the article. Note: even if we tried to be as honest as possible, this is not an independent test, but you can repeat this test yourself anytime at my.goodvisionlive.com.
GoodVision Video Insights is a traffic analysis platform combining automated traffic data retrieval from video with advanced traffic data analytical tools to obtain various traffic performance measures, including the traffic models themselves. Almost all of our current customers on six continents have previous experience with manual traffic counting, and it is often considered a benchmark method. So we have decided to compare the overall user experience of this type of service against automated traffic collection in GoodVision Video Insights. Humans versus machines.
We’ve picked two traffic counting jobs for this contest, both roundabouts.
Job 1: 5-arm roundabout shot by the drone in the United Kingdom, 2 hours, full O-D counts needed, 20 traffic movements
Job 2: 4-arm roundabout from the standard street camera live camera stream, recorded 100 hours in Romania, full O-D counts needed, 12 traffic movements
Part 1: Sending the videos to a traffic counting company in India
COMPANY X: From plenty of companies out there, we picked the medium-sized company (let’s call them Company X), which stated they perform over 50 000 hours of traffic counting annually and provide high accuracy. To start, we were asked to upload the videos to a network shared folder.
Company X then requested the specification of traffic movements — we had to provide screenshots from the video, draw the movements by arrows, and name it one by one. Then we were asked what classes of vehicles we require. We require just for 5–class classification: car, van, truck (including OGV2), bus, motorcycle according to UK Highways Classification Scheme definition. That was not enough.
We had to provide an exact specification of each vehicle class, including the image examples of vehicles from each class so that Company X can do it accurately. … I understand we did it for the first time with Company X, but still … I thought UK scheme requirements should be pretty clear to everyone working in this industry. This “project ramp-up” took us a whole day and four emails.
Security question arose — what about data privacy? How is my data privacy secured with human counters?
. . .
GOODVISION: After the job was specified well enough for Company X, we did the second part of the race: we took the video recordings and uploaded them into the GoodVision Video Insights. (PS: the second job was from the live camera, so if we didn’t need to make the comparison, we could easily connect the live stream to GoodVision and schedule the survey for 100 hours).
Part 2: Waiting for the results
GOODVISION: For Job 1, we received the data from GoodVision 67 minutes after upload. For Job 2, we received the data from GoodVision in 75 minutes. Defining movements for both jobs took us around 10 minutes, and we obtained traffic movement count reports to the mailbox 3 minutes after. The whole procedure took us approximately 1.5 hours for the complete survey, and the traffic reports were on the table.
COMPANY X: Company X promised we would get our results for both jobs in 3 days. Job 1 was offered to deliver in 24 hours for an extra charge, which we refused. On the 3rd day, we emailed Company X to check how things are going and find out we are not getting the results today because of India’s bank holiday. We had to wait another day when the spreadsheets with traffic volumes finally arrived after ~90 hours.
Part 3: We did a configuration mistake, what now?
Here I have to say, and we did this on purpose. Each approach on the roundabout from Job 2 had a “self-exit” movement, which we “forgot” to specify for Company X to count manually.
COMPANY X: We asked Company X what to do, how it goes when an error like this appears. The only option was to re-count the traffic on four more movements (1 movement per each of the four approaches). Luckily they still had the footage on their hard drives (luckily? is your client ok with this?). We had to pay for the new counting. The amount to pay was lower than in the first run, and we had to wait another three business days for updated results.
. . .
GOODVISION: As GoodVision Video Insights video processing always extracts all data from the traffic scene, the missing traffic movements were also collected. We needed to define four new traffic movements on the roundabout and generate a new report. It took us just a couple of minutes, and the traffic volume report was in the mailbox — all with no additional costs.
Part 4: Evaluating the results
We performed a side to side comparison to find out; both methods provided very similar results. Counts from Company X were often lower than the ones from GoodVision. As GoodVision never over-counts volumes on turning movements, it points to missing vehicles in the Company X reports. On the other hand, in several intervals, the counts from Company X were bigger, producing a balanced result of both methods within the accepted accuracy range overall.
How do you evaluate the accuracy of human traffic counts? By other, in-house human counts? And which one do you consider as the ground truth? Yours? Really?
Conclusion — evaluating the experience
Don’t waste your precious time with manual traffic counting. It’s not just about the data turnaround time. Just imagine — all we got from manual traffic counters were spreadsheets with traffic volumes. With an AI-powered software platform like GoodVision Video Insights, you can get way more than that. You get the complete raw data about traffic behaviour, traffic metrics like time-gaps, travel times, speed, exact vehicle traces, saturation flows, which you or your clients — the traffic modellers can utilize for model calibration directly. If you want to enjoy your traffic data collection tasks, GoodVision Video Insights is the solution to go.
The price for human counting grows according to the number of traffic movements, the number of vehicle classes, and traffic density. In contrast, GoodVision always provides a flat price regardless of the scene complexity. We cannot reveal the specific costs of Company X. Still, in general, human counting offers only a bit lower price per hour for the most straightforward traffic scenes, i.e. highways. In contrast, the locations with more traffic movements, or the congested ones, are more expensive than GoodVision.
Speed of delivery
Delivery time of human traffic counters can vary according to the amount of video data provided and can be negatively affected by holidays and staff outages. In contrast, GoodVision delivers data within 1 hour after the video is provided. This might sound fine, but imagine you have a burning deadline and the supplier prolongs the delivery because of XYZ reason. This is STRESS. And our customers want it stress-less.
The #1 benefit the customers are mentioning in their feedback to GoodVision is the stress-less process on their projects thanks to automation and the platform’s ability to scale even for massive projects.
Type of data
Human counters provide static traffic volume reports in Excel spreadsheets. In contrast, GoodVision provides whole raw data in the trajectory format, multipurpose, allowing for repetitive analysis with changing filters. That eliminates the error of specification and allows for multiple other purposes in the future, like obtaining traffic parameters for traffic modelling. Check out the various traffic reports you can receive in GoodVision.
Both methods provide acceptable accuracy. GoodVision video analytics is more sensitive to input quality. It requires you to prepare video footage with consistent quality to process it accurately. GoodVision “replaces eyes”. In contrast, a human counter can sometimes see “behind the corner” and figure out things even if his eyes don’t see them.
Video footage for manual traffic counters shall be sent to them via cloud storage. If the footage is confidential, I believe you’ll need the approval of your client, etc. I’m not sure how the GDPR compliance in the European Union can be met. In contrast, GoodVision Video Insights is GDPR compliant. It uses machines to process the video footage with no third parties having access to it, and it is deleted immediately after processing.
Try GoodVision Video Insights in a Trial
If you feel frustrated or intimidated by all that tech-talk and variety of vendor lock-in single-purpose solutions out there, then GoodVision is the way to go. If you need to process just a few or thousands of hours of traffic videos, Video Insights can handle it. It is built to scale even for your massive projects and always delivers traffic data in 1 hour. Video Insights is a broad ecosystem of capabilities and a one-stop shop for all the traffic analytical tasks the transportation industry requires. Our goal is to remove the stress from your traffic jobs with automation and conserve your time for professional tasks!