Red Light Running Innovation Project
Site visit by Transport for Greater Manchester (TfGM) Road Safety & VivaCity

1. The Challenge: Catalyst for a Development Partne

• Understanding Behavioral Context: TfGM needed visibility into not just if violations occur, but when, under what conditions, and whether they create genuine safety risks for vulnerable road users (VRUs).
• High-Risk Interactions: Identifying red-light running (RLR) trends at complex locations—such as high-risk tram junctions, busy pedestrian crossings, and active travel corridors.
• Targeted Interventions: Enabling data-backed decisions on where enforcement is necessary versus where design, signal timing, or behavioral campaigns are better suited.
• Maximising Existing Investments: Leveraging Greater Manchester’s extensive network of over 400 deployed VivaCity sensors rather than installing standalone, single-purpose infrastructure.

2. Why Not Integrate Directly with the Signal?

• High Cost & Complexity: Wiring directly into legacy signal hardware requires physical cabinet modifications, civils work, and strict signal authority approvals, driving up deployment costs and timelines.
• Purely Visual Approach: By using video analytics to observe the physical illumination of the red signal lamp on the traffic light head, the system operates completely independently of signal control hardware. It also make it highly adaptable to monitor red light running with any type or age of signals.
• Rapid Deployment: Non-intrusive vision-based monitoring eliminates cabinet access dependencies, allowing RLR capabilities to be rapidly deployed across existing sensor estates.
• Highly Accuracy: Both red light detection and stopline crossing detection are done on the same processor with the same clock, so zero risk of clock drift or differences in timing, which matters when you're detecting someone has run a red light by a second or less

3. Hardware Innovation: V3 Multi-View Capabilities

• Multi-View System Architecture: Earlier sensor generations lacked the dual processing capacity required for simultaneous wide-area tracking and focused signal monitoring. The V3 sensor supports up to 3 simultaneous viewpoints (VPs) from a single unit.

• Flexibility of Dual-Viewport or Single Detection:
  • Viewport 1 (Signal Head): Dedicated to monitoring the traffic light signal head.
  • Viewport 2 (Stop Line): Positioned over the carriageway stop line to track vehicles and countline crossings.
  • Single viewport: where both are visible in the field of view
• On-Device Edge AI Processing: Red light detection and vehicle tracking are processed directly on the sensor unit in real time, maintaining strict Privacy-by-Design principles (video is processed and discarded instantly on the edge and not stored).

4. About the Test Junction

• Location: Key arms at the complex junction outside Manchester Piccadilly Station (Fairfield Street / Whitworth Street).
• Context: A busy multimodal hub carrying high volumes of vehicular traffic, buses, cyclists, and pedestrians.
• Proof of Concept Scope (Phase 1): Focuses on establishing real-world accuracy for red-light detection, countline crossings, and correlating RLR occurrences with near-miss interaction datasets.

5. Setting Up Countlines & Defining RLR Events

Technical Setup

Defining an RLR Event

• Our Methodology: An RLR event is triggered when the rear of the vehicle fully crosses the stop line while the red light status is active. This harnesses the benefit of our market-leading 3D footprint detection technique to evaluate the true form of the road user.
• Rationale: Requiring the back of the vehicle to clear the countline ensures the vehicle has fully committed to crossing the junction during a red phase. This filters out minor encroachment while accurately capturing clear red-light runs.

6. Where Next? Wider Rollout

• Phase 1 (H1 2026): Trial validation at Manchester Piccadilly, producing combined CSV datasets for Red Light Running and Near Miss events.
• Phase 2 (H2 2026): Expansion across 10 additional high-priority junctions in Greater Manchester.
• Dashboard Integration: Fully integrating RLR analytics into the central VivaCity Dashboard. Transport planners will be able to filter events by time elapsed after red, vehicle classification, minimum speeds, and view relative percentages of overall signal compliance.
• Safety Integration: Correlating RLR data directly with Near Miss analytics to evaluate how signal non-compliance impacts vulnerable road user safety across the region.

Dashboard site comparison tool:

Sensor 399 is the standout for highest number of RLR incidents:

Diving further into the data at this site allows us to see peak times this is happening - 2:30 - could this be school traffic? Allows TFGM to more effectively deploy spot checks with Police.

Speed is also an issue - in a 7-day period the RLR recorded 56 incidents at 25-30mph which could have very serious consequences if there is a pedestrian entering the crossing on the other side of the junction.

General features:

Classified counts: 
Ped trends at the crossing where RlR behaviour is occurring:

Then Classified counts in the carriage way at the same location for vehicles and cycles:
Very busy corridor - averaging 11k - 13k of vehicles a day

Busy all day, but RLR behaviour particularly at 2-4pm.

Speed isn’t particularly a problem: but interesting to show all the same.