Solution
Turn traffic camera feeds into actionable Mobility Analytics
Track, classify, and analyze multi-modal traffic and pedestrian movement in real time. Deploy edge-processed and GDPR-compliant analytics on your existing CCTV infrastructure without sending a single byte of video data to the cloud.
98% + Vehicle detection accuracy in challenging outdoor environments
99.8% Pedestrian detection reliability for live infrastructure actuation
100% GDPR compliance via local, air-gapped edge processing
Integration with VMS software like Milestone’s Xprotect and Genetec
Works with existing IP cameras and legacy analogue CCTV infrastructure
Integration with traffic controllers (Swarco, Yunex etc.)
GDPR-compliant by design — real-time face and license plate blurring and no biometric data
Multi-Modal Traffic Counting & Vehicle Classification
Local and regional authorities need accurate utilization data to justify infrastructure investments. MarshallAI replaces manual counters and delicate legacy loops with automated visual intelligence that functions in heavy rain, thick fog, and sub-zero Nordic winter conditions.
Our platform does not just count traffic; it categorizes the composition of your roadways simultaneously.
Granular Classification: Differentiate between passenger cars, commercial delivery vans, heavy freight trucks, transit buses, cyclists, and pedestrians.
Turning Movement Counts (TMC): Automate complex intersection tracking to map exact vehicle pathways and volumes.
Asset & Lane Utilization: Identify bottleneck areas, unauthorized bus lane usage, and structural queue delays in real time.
Crowd Analytics & Pedestrian Desire Lines
For public safety, venue management, and smart city planners, understanding human movement is vital to optimizing spaces and protecting citizens.
MarshallAI turns any standard IP camera into an intelligent sensor that measures spatial dynamics without compromising individual privacy.
Pedestrian Desire Lines: Map the actual paths people take across public squares, transit hubs, or commercial spaces, exposing layout inefficiencies.
Real-Time Crowd Density: Monitor occupancy levels in defined zones to prevent overcrowding and trigger automated safety alerts.
Zonal Dwell Times: Calculate exactly how long visitors linger in front of specific infrastructure, exhibits, or transit platforms.
Engineered for Privacy: 100% Air-Gapped GDPR Compliance
Public sector procurement demands strict data sovereignty. Traditional cloud-based systems stream live, identifiable video externally, creating immense regulatory liability.
MarshallAI completely de-risks your deployment through our local inference engine and dedicated privacy nodes.
Real-time privacy: our platform utilizes a configurable real time blurring node. This node automatically masks all human faces and vehicle license plates in before the video feed ever reaches your control room or debugging monitors. Operators see the spatial environment and the tracking data, but no personally identifiable information (PII) is ever visible.
Edge processing: completely offline processing. The AFSA platform creates anonymized numerical statistics locally and optionally destroys the unmasked footage instantly at the edge.
Native Infrastructure Integration
A software solution is only useful if it talks to your physical assets. MarshallAI bridges the gap between modern AI models and legacy hardware platforms through native protocol support.
| Protocol / System | Integration Type | Operational Benefit |
|---|---|---|
| DATEX II / RSMP / OCIT | European Traffic Protocols | Connects directly to Swarco, Yunex, and Nordic roadside controllers to actuate live signals and export standardized traffic data. |
| Milestone XProtect / Genetec | Video Management Systems | Overlays real-time mobility data directly onto your existing security dashboard. |
| JSON / REST API | Data Infrastructure | Exports automated traffic statistics directly to municipal dashboards and ERPs. |
Case studies
-
Actuating Physical Infrastructure: How MarshallAI Achieved 99.8% Reliability in Harsh ConditionsDiscover how MarshallAI’s edge vision flawlessly tracks complex objects in extreme conditions, achieving a 99.8% pedestrian detection reliability rate while safely triggering physical infrastructure controllers in real-time.
Smart cities Traffic management
-
Smarter intersections, fewer emissions: MarshallAI and AI4CitiesMarshallAI is proven in critical systems and continuous operation. Deployed in Helsinki and Paris, our deep learning platform autonomously controlled live infrastructure to reduce emissions by up to 8%.
Smart cities Traffic management
-
Upgrading Legacy Infrastructure: Crowd Analytics & Security Without Cloud DependencyLearn how we utilized Särkänniemi theme park’s existing camera infrastructure to process live video on a closed-loop local area network – enabling security and crowd flow analytics while maintaining privacy.
Security Smart cities
Answers about mobility analytics
Can our team still view the live video feeds for monitoring or system debugging without violating GDPR?
Yes. The AFSA platform includes a native, real-time blurring node designed specifically for European privacy mandates. Before any video stream is transmitted to an operator’s dashboard or debugging view, the platform automatically applies a privacy mask over all human faces and vehicle registration plates. This processing happens locally. It allows your operations team to monitor traffic flows and verify system health visually, while guaranteeing that no personally identifiable information (PII) is ever exposed, recorded, or transmitted.
Do we need to replace our existing traffic and security cameras to use your analytics?
No. MarshallAI is entirely hardware-agnostic and built to leverage your existing infrastructure. The platform natively integrates with standard IP security cameras via RTSP or ONVIF streaming protocols, as well as legacy analog feeds. This means you can transform your current municipal CCTV network into an intelligent sensor grid without the capital expense or construction disruptions of installing new cameras or digging up roadways for induction loops.
How reliable is the system’s detection accuracy during winters, heavy fog, or dark conditions?
The AFSA engine is engineered and field-tested in Finland to withstand extreme northern climates. In controlled industrial environments, the system achieves 99.9995% accuracy, and it maintains over 98% tracking and classification accuracy in harsh, outdoor real-world conditions like heavy snow, thick fog, and low-light winter environments. Using heated and properly placed cameras we can minimize weather effects.
We use Swarco and Yunex controllers. Can MarshallAI actuate our existing traffic signals directly?
Yes. MarshallAI bridges the gap between AI visual tracking and European roadside hardware. The system features native support for European transportation protocols like RSMP (the dominant standard across Scandinavia) and OCIT (Central Europe), alongside DATEX II for regional traffic data exchange. Rather than just collecting passive statistics, the system communicates directly with your existing traffic controllers to trigger, live signal actuation—allowing you to eliminate “dead seconds” at intersections dynamically.
Can the platform track and differentiate between micro-mobility users, like cyclists and e-scooters, or does it just count “pedestrians”?
Our platform provides granular, multi-modal classification. It does not generalize vulnerable road users (VRUs). The AFSA models are trained to clearly differentiate passenger cars, delivery vans, heavy freight trucks, transit buses, cyclists, e-scooter riders, and pedestrians simultaneously. This high-fidelity classification allows us to generate accurate Turning Movement Counts (TMC) and measure precise Modal Share data.
What is the turnaround time for training and deploying the system?
MarshallAI is optimized for rapid deployment. Our deployment benchmarks prove that a custom, highly accurate functional model can be built, trained, and deployed from as few as 77 reference images in under 100 minutes. But to get the most accuracy model we usually refine the model during several months to capture different lighting and weather conditions.
