On every construction site, the amount of waste produced is important. It should therefore be managed to minimize the impact on the environment, and reduce costs.
Today, construction companies usually sort the waste into different containers, which are then sold to recycling companies. However, the sorting of waste on site is often far from perfect, which means that the content of the containers is often not pure enough to be recycled, and the container will have to be sold at a lower price.
Our client, Besix is a leading Belgian construction company operating worldwide. To improve the waste management on their construction sites, Besix was looking for a solution to automatically monitor the quality of the sorting in the containers, and alert the construction site manager in case of sorting errors.
To solve this problem, Sagacify built an AI solution that classifies the content of containers based on images taken from cameras on the construction site.
During the project, construction sites were not opened due to COVID and we only had a handful of data to train the model. It was an excellent opportunity to look for solutions to cope with the lack of data!
To solve the problem of low annotated data, Sagacify used a self-supervised learning approach which helps maximizing the model performances on a small dataset. The data itself was connected with an ordinary camera fixed on a crane, at several construction sites. The images were later annotated by Besix collaborators.
The solution is able to detect the containers on construction site images, classify their content and assess the proportion of contaminants per container, with a high level of accuracy
It automatically sends an alert to the responsible of the construction site when contaminants are detected.