X-AID™ Video Incident Detection for Keilaniemi Tunnel, Finland

Project Overview

Telegra delivered the X-AID™ video-based incident detection system for the Keilaniemi Tunnel in Finland.

The system supports automatic detection of stopped vehicles, wrong-way vehicles, and slow vehicles, helping tunnel operators detect incidents quickly and initiate appropriate incident management actions such as lane closure, tunnel closure, driver warning, and accident response.

X-AID™ analyzes video feeds from dedicated tunnel cameras and reports incidents to the tunnel management center. In addition to incident detection, the system provides average traffic speed information for each lane.

Client Challenge

Tunnel incidents can have severe consequences, including human casualties and major material damage. Tunnel authorities must detect incidents quickly and initiate immediate safety responses to prevent escalation.

Fintraffic required reliable automatic detection of stopped vehicles, wrong-way vehicles, and slow vehicles in tunnel conditions.

The main operational challenges included:

• automatic detection of stopped vehicles,
• automatic detection of wrong-way vehicles,
• automatic detection of slow vehicles,
• fast incident awareness for tunnel operators,
• reduction of missed detections,
• reduction of false alarms,
• support for incident management strategies such as lane closure, tunnel closure, driver warning, and accident response,
• reliable performance in challenging tunnel camera conditions,
• short and effective tuning period compared with traditional video detection systems.

Before X-AID™, traditional video detection technologies required long tuning periods and relied on sensitivity trade-offs between detection rate and false alarm rate. Reducing false alarms often affected detection rate, and increasing detection rate often increased false alarms.

Telegra Solution

Telegra implemented X-AID™, an AI- and machine-learning-based video incident detection system.

X-AID™ learns from real-life tunnel conditions and fine-tunes detection performance to improve incident detection and reduce false alarms. Unlike traditional systems that rely mainly on detecting changes between consecutive video frames, X-AID™ uses machine-learning methods to detect vehicles, separate objects, and improve incident localization.

The solution included:

• X-AID™ video-based incident detection,
• automatic detection of stopped vehicles,
• automatic detection of wrong-way vehicles,
• automatic detection of slow vehicles,
• average traffic speed measurement per lane,
• analysis of video feeds from dedicated tunnel cameras,
• reporting of incidents to the tunnel management center,
• AI/ML-based fine-tuning in real tunnel conditions,
• centralized redundant system architecture,
• hot-spare N+1 redundancy,
• support for reliable operation in challenging tunnel video conditions.

Tunnel Environment

The Keilaniemi Tunnel is a two-way tunnel in Finland.

Key tunnel characteristics include:

• approximately 495 m tunnel length,
• main tunnel cross-section with three lanes plus an emergency lane in each direction,
• open-space height of approximately 5.2 m,
• total tunnel height of approximately 7.0 m,
• 20 cameras in the eastern tunnel tube dedicated to video incident detection,
• 18 cameras in the western tunnel tube dedicated to video incident detection,
• detection areas covering both tunnels in the driving direction from inside the tunnel toward the tunnel exit and the edge of the Otasolm bridge,
• logical areas corresponding to fire exhaust areas,
• each tunnel pipe divided into 7 logical areas.

Each detection area consists of up to two lanes and three logical areas, allowing incident localization by lane and logical tunnel area.

Detected Events

The X-AID™ system in the Keilaniemi Tunnel detects:

• stopped vehicles,
• wrong-way vehicles,
• slow vehicles.

The system also provides:

• average traffic speed per lane,
• incident reports to the tunnel management center,
• lane-based and logical-area-based incident localization.

Detection Challenges

The Keilaniemi Tunnel presented several technical challenges for video-based incident detection.

These included:

• low camera height,
• vehicle occlusions in dense traffic,
• environmental conditions such as wet road surfaces, water puddles, snow, dirty lenses, and low sunlight,
• shadows and illumination changes,
• non-uniform tunnel lighting,
• camera noise and low-visibility conditions,
• need for precise vehicle separation,
• need for reliable per-lane and per-logical-area detection.

Low Camera Height & Vehicle Separation

Cameras in the Keilaniemi Tunnel are mounted at a height of approximately 5.5 m.

This camera height can produce images where vehicles, especially those farther from the camera, are partially occluded by other vehicles in dense traffic. In such conditions, gaps between consecutive vehicles may not be clearly visible because of perspective transformation.

X-AID™ addresses this challenge using a machine-learning approach. The system observes each video frame and identifies vehicle bounding boxes, rather than relying only on changes between consecutive frames.

This supports:

• improved vehicle separation,
• better understanding of dense traffic scenes,
• more reliable incident detection,
• more accurate traffic statistics,
• precise incident localization by lane and logical area.

Environmental and Lighting Challenges

Finnish weather conditions create additional challenges for automatic video incident detection.

The system must operate in conditions that may include:

• wet road surfaces,
• water puddles near tunnel portals,
• snow during winter,
• snow falling from vehicle roofs inside the tunnel,
• dirty camera lenses caused by water particles and traffic spray,
• low-positioned sun penetrating deep into the tunnel,
• shadows within the camera field of view,
• changing tunnel lighting regimes.

X-AID™ addresses these conditions through machine-learning-based vehicle detection and object recognition. The system is trained to detect vehicles and discard irrelevant shapes caused by weather, shadows, reflections, lighting changes, and background variations.

Centralized Redundant System Architecture

X-AID™ is a software solution installed on commonly available workstations or servers, referred to as X-AID™ analyzers.

Video streams from cameras are delivered to the X-AID™ analyzers over the network. This architecture makes hardware maintenance easier because processing hardware can be located centrally.

The Keilaniemi deployment supports hot-spare N+1 redundancy, meaning that the system is not affected by a single analyzer failure.

The centralized redundant architecture supports:

• centralized video processing,
• easier maintenance of analyzer hardware,
• use of standard workstation or server hardware,
• hot-spare redundancy,
• continued system operation in case of single analyzer failure,
• scalable deployment for tunnel video analytics.

Operational Impact

X-AID™ proved to be a reliable video incident detection solution in the Keilaniemi Tunnel.

Documented outcomes include:

• detection rate above 95%,
• false alarm rate of 0.17 false alarms per camera per day over a seven-month period,
• proven daily operational use in the Keilaniemi Tunnel over the last 7 years,
• significant improvement compared with the client’s previous traditional video detection systems,
• shorter tuning period compared with previous experience,
• reliable detection of stopped, wrong-way, and slow vehicles,
• average speed measurement,
• improved support for tunnel management center operators,
• improved ability to respond promptly to potentially life-threatening tunnel incidents,
• AI/ML-based fine-tuning in real-life tunnel conditions,
• central redundant architecture with hot-spare N+1 redundancy.

Related Software Solution Areas

• Incident Detection & Incident Management Workflows,
• Tunnel Control & SCADA Supervision Systems,
• Traffic Management Center (TMC) Platforms,
• ITS Integration & Interoperability,
• Traffic Analytics, Reporting & Operational Intelligence.

Download Case Study

Download the original PDF case study:

Finland - X-AID™ Video Incident Detection – Keilaniemi Tunnel

Frequently Asked Questions

What is the Keilaniemi Tunnel X-AID case study about?

It describes Telegra’s implementation of the X-AID™ video-based incident detection system in the Keilaniemi Tunnel in Finland.

What type of system was delivered?

Telegra delivered X-AID™, an AI- and machine-learning-based video incident detection system for automatic tunnel incident detection and traffic-speed monitoring.

What incidents does X-AID™ detect in the Keilaniemi Tunnel?

X-AID™ detects stopped vehicles, wrong-way vehicles, and slow vehicles. It also provides average traffic speed information for each lane.

How many cameras are used for video incident detection?

The system uses video feeds from 20 cameras in the eastern tunnel tube and 18 cameras in the western tunnel tube dedicated to video incident detection.

What detection performance was achieved?

The case study reports an incident detection rate above 95% and a false alarm rate of 0.17 false alarms per camera per day over a seven-month period.

Why was the Keilaniemi Tunnel environment challenging?

The tunnel presented challenges such as low camera height, vehicle occlusions, wet road surfaces, snow, dirty lenses, low sunlight, shadows, lighting regime changes, low visibility, and camera noise.

Does the system support redundancy?

Yes. The X-AID™ architecture supports a centralized redundant system with hot-spare N+1 redundancy, so the system is not affected by a single analyzer failure.

Does this case study support Telegra’s tunnel video incident detection positioning?

Yes. The project demonstrates Telegra’s capability to deliver AI-based video incident detection for tunnel environments, including stopped vehicle detection, wrong-way detection, slow vehicle detection, average speed monitoring, low false alarm performance, and redundant architecture.