Detecting hazardous physical activity

Using human action detection to monitor safety in hazardous or physically demanding situations, such as escalator use and offshore wind turbine work


The accurate classification of human movement from automated CCTV monitoring of customers in retail environments and of workers performing strenuous tasks has the potential to provide early warning for hazards and increase their protection. Cutting-edge deep learning technologies in combination with the technique know as ‘path signatures’, will allow for the development of domain-specific classifications of different action patterns to improve the safety of these particular contexts.

Explaining the science

This project's work draws on fundamental research being conducted at the Turing in to the use of 'path signatures' in capturing complex data streams.

See the related project ‘Capturing complex data streams’ for more information about path signatures. 

Project aims

To economically and effectively represent the moving human body in action detection. This will promote mathematical innovation in the understanding of evolving streamed data and its integration into the existing state of the art of action detection. 

Working in collaboration with industry partners and the Health and Safety Executive, the aim is to apply this work to make useful tools for monitoring of hazardous or physically demanding tasks using CCTV, and so contribute to safety in the public spaces and the workplace. 

This research will draw upon the cutting-edge 'path signature' method, in conjunction with pose detection, and other machine learning techniques and algorithms, to progress the challenge of human movement recognition and detection with the focus on applications in health and safety. The potential applications range from automated monitoring of the customers in retail environments to early detection of hazardous actions of remotely located workers when performing strenuous tasks.

Pilot research has achieved state-of-the-art results in standard action datasets.


The work in this project is focusing on two example application situations: customer use of escalators in retail environments and workers performing strenuous tasks on offshore wind turbines. However the techniques being developed have the potential to be applied to any industry where there are people working or acting in ways that could be hazardous, and where there could be CCTV monitoring available to help.


Researchers and collaborators