Unattended object detection at train stations

KickstartAI X NS - AI-supported surveillance

Together with NS (Dutch Railways) we are working on detecting unattended objects on railway platforms. By linking existing camera infrastructure to AI algorithms, the system helps control room personnel identify potential threats. Following a successful technical proof of concept we are refining and implementing it.

Harnessing AI to make train station safer

NS has access to a vast network of cameras across its stations for security purposes. The control room plays a crucial role in reacting to potential safety threats, including unattended objects. But the human eye can not watch everything at the same time. That is why we collaborated on an AI-powered system that automatically detects unattended objects.

Goals

1. Reduce the workload of the control room 

2. Improve safety and perceived safety on NS stations

3. Enhance NS's expertise in using AI 

Approach

When building the AI system, we found that a key feature would be its real-time processing capability. The AI algorithms analyze live camera footage, enabling immediate identification of unattended objects. Real-time analysis allows for a swift human response, timely eliminating potential threats.

We organized a challenge to develop a proof of concept with data scientists from NS, KickstartAI, and our partner companies. The team addressed challenges such as anonymization, object recognition, object tracking and video preprocessing. 

Key features 

Real-time detection 

Analyzes live camera feeds to identify unattended objects.

Scalability

Designed to handle an increasing number of camera feeds as the system expands.

Privacy measures

Incorporates privacy-preserving measures to protect passengers and personnel.

Making Dutch railway stations safer

The AI-supported unattended object detection system is a significant step forward in security measures. Together with NS, we are now working on bringing it into production. By automating threat detection, the control room will be more efficient and response times will be reduced. This improves overall security at train stations in the Netherlands.

Results

The proof of concept has proven its effectiveness on video streams. Once being brought to production, the system is expected to:

• Reduce the time required for control room personnel to assess

• Increase the detection rate of unattended objects

• Shorten the time to alert for unattended objects