loading

Business Impact

  • 90%

    Accuracy

  • 10-15 FPS

    Inference engine running

Customer Key Facts

  • Location : North America
  • Industry : Transportation

Problem Context

In an effort to keep employees, consumers’ goods and communities safe, one of the largest railways is undertaking large-scale initiatives for situational awareness around train tracks. The customer relies on its 17,000 trained crew members to manually identify flags and signals as well as humans and their capacity (i.e. crew or trespassers), interpret their meaning and their association with their respective track, and take appropriate actions according to the Rail Operating Rules. However, the possibility of accidents as a result of poor human judgment poses a real concern for employees and civilians.

Challenges

 

  • Delayed response time – Crew members must manually identify upcoming flags and signals
  • Trespasser Hotspots – Difficulty in detecting trespasser hotspots, leading to unwanted business losses (i.e. theft of cargo goods)
  • Lack of Connectivity – No solution present for connecting CNR offices to the edge device for checking the deployed system and hardware conditions
  • Hardware Limitations – The edge device to be deployed had limited memory to run multiple classification and detection models
  • Data Insufficiency – The images of signals in bad lighting conditions and obstructed views, etc. results in poor prediction accuracy
Challenges

Technologies Used

Google AutoML
Compute Engine
Vision API
TensorFlow

Autonomously Detecting Hazards & Objects Near and Around Railway Tracks With Intelligent Video Analytics

Solution

Quantiphi leveraged Google’s open-source TensorFlow library, CloudML Engine, and other Google Cloud products to develop a custom deep learning solution to perform intelligent video analytics and provide real-time assistance to crew members by autonomously detecting objects and people near and around its 20,000-mile network of railway tracks. Machine learning models were designed to infer the signals, flags, and the crew members present on the tracks as well as store this data in real-time on a web dashboard. The Trespasser Detection System helped towards identifying trespasser hotspots, enabling crew members to take actions to mitigate regions where trespassers are often prevalent. The Crew Assisted Monitoring System allowed for faster and more accurate identification of the concerned objects in frame and quicker alerts sent for taking actions.

Result

  • Better detection of trespasser hotspots, reducing theft
  • Increased monitoring and safety for employees, consumer goods, and communities

Looking for similar project?

Let's Talk

Get your digital transformation started

Let's Talk