Computer Vision Method for Safety Hazards Detection, Referencing, and Tracking

Case ID:
UA24-037
Invention:

This invention is a method for automatic identification and real-time mapping of safety hazards and individuals within underground spaces, such as mines. Using a combination of visual, Near-Infrared (NIR), and LiDAR imaging technologies with a robust, low-cost, machine learning-based algorithm, this innovation captures data and geo-references and categorizes it before transmitting it to a centralized hub for mapping. One of the notable features of this invention is its reliance on collaborative and federated learning, which enables fine-tuning of the machine learning model's parameters and collectively improves the model's ability to learn from the data provided by all users. This technology is adaptable and can be deployed via a user-friendly Human-Machine Interface (HMI) supporting real-time retrieval for localization and rescue missions. While initially designed for mobile-device apps, it can be integrated into various infrastructure setups, including surveillance cameras and mobile robotic platforms. Ultimately, this innovation revolutionizes mining safety by combining distributed information gathering, machine learning, remote analysis, and workflow optimization to effectively assess and mitigate risks effectively. Mining operators can utilize their smart devices to gather data, which is then used to provide live risk assessments. 

Background: 
Currently, mining safety operations rely on manual hazard identification and mapping processes, which are time-consuming, error-prone, and often subject to human limitations. These processes can lead to delays in responding to emerging hazards, potentially endangering workers’ lives. Moreover, existing technologies such as surveillance cameras and manual data collection lack the real-time capabilities and adaptive learning capabilities that this innovation offers. 

Applications: 

  • Mining and construction
  • Tunneling and subway maintenance
  • Risk assessment in underground environments


Advantages: 

  • Enhanced safety and efficiency
  • Real-time adaptive response to environmental change
  • Enables immediate action based on swift and accurate risk assessment
  • Collaborative and federated learning approach to improve machine learning model
  • Streamlines workflows
  • Minimizes risk of human error
  • Can be integrated into various infrastructures
Patent Information:
Contact For More Information:
Tariq Ahmed
Sr Licensing Manager, College of Engineering
The University of Arizona
tariqa@tla.arizona.edu
Lead Inventor(s):
Maria Nathalie Risso
Angelina Anani
Pedro Lopez-Vidaurre
Keywords: