AI-Supported Tunnel Surveillance and Anomaly Detection

Project Description

This project aimed at helping in monitoring car tunnels by analyzing a stream of sensor data in real time.

Tunnels generate thousands of data points each second. Multiple sensors at different locations measure carbon monoxide. Every light has a sensor that signals if it is turned on or off. Multiple traffic control signs can be changed and their functionality constantly needs to be monitored. On top of that cameras are positioned at crucial positions for tunnel operators to observe what is happening in a tunnel.

All of that data holds valuable information about the situation in the tunnel. However because of the sheer amount of data, most of this information is unused and we rely on experienced operators to make the decision which parts of the tunnel to pay attention to at any given moment. In order to be able to quickly diverge an operators focus to dangerous situations, we designed and implemented a Machine Learning assisted solution that analyzes the data stream in real time and finds anomalies in the data stream the moment they occur.

Project Outcome

We successfully implemented a real time analysis model which could recognize anomalies in the data stream. We routed every sensor update through a message queue (RabbitMQ) to build a global state of all tunnel sensors including historic values which could then be analyzed by a multivariate Time Series Model.