VolkerFitzpatrick is amongst the leading engineering and construction companies in the UK. The organisation has projects in a wide range of market sectors, including commercial, industrial, education, rail, aviation, waste, and energy. Their question: how can artificial intelligence be employed on construction sites to increase productivity, reduce safety incidents and gain better management controls?
VolkerFitzpatrick’s objective asked for a flexible solution that could answer a wide range of questions. Are deliveries being made on time? How many people are on the construction site at any given time? Is everyone wearing the required protective equipment? The solution we selected is called ‘Hyrde Air’: a bespoke technology that gathers data from still images using artificial intelligence. The system recognises anything from equipment such as bulldozers and forklifts, to aggregate like sand or gravel, to people and the protective equipment they are wearing. Through deep learning, the system’s accuracy is remarkably high at 99.9%.
The possible applications of the data are limitless. Aside from having a real-time dashboard displaying a project’s status, inefficiencies can automatically be detected. Examples include building materials stored at the wrong location on site or the number of hours certain equipment is found at a section of the construction site. Hyrde Air can also detect if people are wearing proper PPE and whether authorised walkways on site are being respected. Moreover, various tasks can be automated based on the data. For example, Hyrde Air can notify a security company when a construction site is accessed after hours. And by identifying the color of hardhats, the system can also detect the location on site of foremen or other staff members.
Hyrde Air protects privacy by design and is fully GDPR compliant. It only captures low resolution images on which private data such as facial features or license plates are not recognisable. Moreover, ‘blurring’ and ‘masking’ modules make sure that certain images such as passing pedestrians or license plates are not captured. After processing, image files are automatically deleted from the cloud as well.
“VolkerFitzpatrick was looking for a way to make its construction sites even safer and more productive through artificial intelligence. Hyrde has delivered a remarkable solution that has already shown promising results.”
After carefully taking stock of VolkerFitzpatrick’s requirements, we considered various solutions. Hyrde Air was found to be fit for purpose because of its extreme flexibility. The system can be used to achieve a wide range of goals, from optimising productivity to improving safety and security. Moreover, a large advantage of Hyrde Air is that it is ‘camera agnostic’: it works with any camera system. As a result, hardware already in place can be used, making roll-out very fast.
The construction site for the Feltham train station was selected as a suitable testing ground. Two PTZ revolving cameras were placed at strategic points at the site. By configuring ten pre-sets on each camera, we realised coverage of a large area. Next, a library of equipment and other objects was created. This process involves manually labelling each picture with the objects it contained. After a set number of images are captured though, the deep learning algorithms take over and we can achieve 99.9% accuracy. This means that the manual labelling will be reduced when Hyrde Air is rolled out to other construction sites, as the existing library of images can be used. Only new equipment will require labelling.
All data is collected in our data platform: EKCO. In this case, our client requested to have all the data to do their own processing and analysis, so we set up a server for VolkerFitzpatrick and delivered all the raw data through EKCO.