Accuracy of traffic counting systems: the Mobility Lab benchmark
The accuracy of a traffic counting system determines how reliable the data is for traffic planning, funding applications and infrastructure investment. A count with systematic errors leads to measures that are incorrectly scaled – from traffic light timing to cycle path expansion.
DataCollect employs two complementary measurement methods: the AI video detection system ARGOS and the radar-based SDR side-view radar. The ARGOS AI video detection system achieved an accuracy of over 95% in the Mobility Lab test. Which method delivers the highest accuracy depends on the measurement task – this page explains how accuracy is measured and which system provides the most reliable data for which applications.
How is the accuracy of counting systems measured?
Reference measurement against a verified ground truth
The accuracy of a counting system is determined by comparing the automatically recorded values with a verified reference (ground truth), such as a manual check count. The comparison involves vehicle counts, their classification into object or length classes and – at complex junctions – turning behaviour. Only this comparison makes it possible to make a robust statement about accuracy.
The Mobility Lab test
In the Mobility Lab test, the ARGOS AI video detection system was tested against a reference under real-world traffic conditions. The result: an accuracy of over 95 per cent. This figure was also confirmed with an average daily traffic (DTV) of 70,000 to 100,000 vehicles – that is, under high traffic density, where simpler systems typically lose accuracy.
Over 95 per cent accuracy in the Mobility Lab test – also proven with 70,000 to 100,000 vehicles per day (ADT). The analysis takes place directly on the device (edge AI), ensuring that reliable data is available without any delay.
Factors influencing accuracy
- Detection geometry: Cross-section (a single line) or entire traffic area including junctions
- Level of classification: number of distinguishable object or length classes
- Obstruction and traffic density: high DTV, congestion, dense traffic flow
- Multimodality: Detection of pedestrians, bicycles and other modes of transport
- Ability to retrospectively correct incorrectly delineated counting zones
Accuracy of ARGOS AI video detection
The ARGOS AI video detection system analyses traffic directly within the device (edge AI, on-board). It captures entire traffic areas, including junctions, turning relationships and roundabouts, and distinguishes between 11 object classes – including pedestrians and cyclists. As a result, ARGOS delivers high accuracy, particularly at complex junctions and in multimodal traffic, where a simple cross-sectional measurement reaches its limits.
- Over 95 per cent accuracy in the Mobility Lab test, even at traffic volumes of 70,000–100,000 vehicles per day
- 11 object classes, including pedestrians and cyclists
- Edge AI: on-board analysis, reliable data with no delay
- Capture of turning patterns and source/destination traffic at junctions
- Deployment at junctions, roundabouts and arterial roads (multi-lane roads)
Recounting is a unique feature of ARGOS: In the cloud myTrafficData 2, counting zones and travel relationships can be adjusted even after the measurement has been completed, and the data can be recalculated retrospectively for up to 90 days – without the need for a new measurement campaign. A counting zone that was initially set incorrectly can thus be corrected, rather than repeating the measurement.
Recounting also serves to answer further questions based on a single measurement: the heatmaps show where road users are actually moving; the measurement fields can then be optimally configured. For example, the same measurement can be used to determine both the origin-destination traffic at a junction and the crossing of cyclists via the footpath.
ARGOS detects traffic via an on-board camera and stores only anonymised object data. No personal video data is transmitted to the cloud. Actual video recording is not necessary with ARGOS.
Accuracy of the SDR side radar at the road cross-section
The SDR side-view radar measures across the carriageway cross-section and records vehicle counts, speeds and length classes. It covers up to 3 lanes and classifies vehicles into 4, 6 or more length classes. For simple cross-sections in mobile or continuous operation, the radar provides reliable data without the need for a camera.
- Cross-section measurement of volume, speed and length classes
- Detection of up to 3 lanes
- Classification into 4, 6 or more length classes
- Variants: SDR traffic (bidirectional) and SDR Dual lane (bidirectional and unidirectional)
- Both variants can be used in mobile mode for short-term measurements and in permanent operation
- No personally identifiable data
Scope: What the DSD speed display records
The DSD speed display (smiley display) is not a counting system with object classification. It displays the measured speed (recording from 3 km/h) and promotes road safety. From the DSD smart configuration level onwards, it optionally records anonymised traffic data – volume and speed, including a timestamp – but without object classification. A timestamp is also always provided by ARGOS and SDR; the only distinguishing feature is object classification. For classified traffic counting, ARGOS (AI video) or the SDR side-view radar.
Which technology delivers the highest accuracy for which application?
| Criterion | AI video (ARGOS) | Radar (SDR) |
|---|---|---|
| Detection principle | Camera + edge AI (on-board) | Side radar at the cross-section |
| Object/length classes | 11 object classes | 4 or 6+ length classes |
| Pedestrians | Yes (differentiated) | No |
| Bicycles | Yes (differentiated) | Limited (cross-section) |
| Complex junctions/turning vehicles | Yes (origin/destination traffic) | Not technically feasible |
| Proof of accuracy | over 95% (Mobility Lab test) | verified by testing carried out by external bodies as part of tenders and trial operations, as well as our own reference measurements |
| Retrospective correction | Recounting up to 90 days | Not provided for |
| GDPR | Only anonymised object data | No personally identifiable data |
| Typical applications | Junctions, roundabouts, arterial roads (multi-lane roads) | Simple cross-sections in mobile or continuous operation |
DataCollect provides both methods from a single source and, adopting a technology-neutral approach, recommends the system best suited to the measurement task: AI video ARGOS for complex junctions and multimodal traffic, and the SDR side-view radar for simple cross-sections in mobile or continuous operation. Both systems run on one of the available myTrafficData cloud solutions – the SDR via myTrafficData, ARGOS via myTrafficData 2.
How to determine the appropriate accuracy requirement
Defining the measurement task
First, clarify what is to be counted: pure volume and speed data at the cross-section, or a differentiated count including object classes, turning behaviour and multimodal traffic.
Assigning methods
For simple cross-sections in mobile or continuous operation, the SDR side-view radar is the cost-effective choice. For complex junctions, pedestrian and cycle counting, and high traffic densities, the ARGOS AI video detection system is the method of choice – at junctions and roundabouts, cross-section measurement using the SDR is technically not possible.
Analysis and post-correction
Both systems run via one of the available myTrafficData cloud solutions – the SDR via myTrafficData, ARGOS via myTrafficData 2. With ARGOS, counting zones can be adjusted retrospectively for up to 90 days using the recounting function – this ensures data quality even if the requirements change after measurement has begun. The data from each system can also be imported directly into the users’ higher-level traffic data management systems via FTP transfer or API.
Frequently Asked Questions
Accuracy is determined by comparing the automatically recorded values with a verified reference (ground truth), such as a manual control count. The comparison covers vehicle counts, the assignment to object or length classes, and – in the case of complex junctions – turning patterns. Only this comparison allows for a reliable statement regarding accuracy.
ARGOS achieved an accuracy of over 95 per cent in the Mobility Lab test. This figure was also confirmed under conditions of average daily traffic (ADT) ranging from 70,000 to 100,000 vehicles. The analysis takes place directly within the device (edge AI), and ARGOS distinguishes between 11 object classes, including pedestrians and bicycles.
The SDR side radar measures across the carriageway cross-section and records vehicle counts, speeds and length classes across up to three lanes. It classifies 4, 6 or more length classes and provides reliable data for simple cross-sections in mobile or continuous operation without the use of a camera. Its high levels of accuracy have been verified by external bodies as part of tenders and trial operations, as well as by our own reference measurements.
The DSD speed display shows the measured speed (measured from 3 km/h) and, from the DSD smart configuration level onwards, optionally records anonymised traffic data – including volume, speed and a timestamp. However, it does not perform any object classification; a timestamp is also output by ARGOS and SDR. ARGOS or the SDR side radar are responsible for classified traffic counting.
Recounting is a unique feature of ARGOS: in the myTrafficData 2 cloud, counting zones and travel relationships can be adjusted even after the survey has been completed, and the data can be recalculated retrospectively for up to 90 days – without the need for a new survey campaign. Incorrectly defined counting zones can thus be corrected without having to repeat the measurement. Furthermore, the same measurement can be used to answer further questions: Heatmaps show where road users are moving, enabling measurement zones to be positioned optimally – for example, for origin-destination traffic at a junction or cyclists crossing the pavement.
For complex junctions, roundabouts and multimodal traffic, the ARGOS AI video detection system delivers the highest accuracy because it captures entire traffic areas, including turning patterns, and distinguishes between 11 object classes. The SDR side radar is designed for simple cross-sections and cannot technically be used at junctions or roundabouts. DataCollect offers both methods and recommends the most suitable system on a technology-neutral basis.
Further reading on the comparison of both measurement principles: Traffic counting with radar or AI video.
Are you planning a traffic count and would like to know which method is best suited to your measurement task?