The challenge
A workplace hygiene solutions company asked AI Superior for a system that could identify on its own when an area needed cleaning, reducing the need for human intervention.
The system had to recognise objects and tell clean areas from dirty or messy ones. It also had to detect people in these areas while respecting privacy, so that images containing people were not processed or stored.
Our solution
The project ran in two phases. In the first, we developed an object detection model that identifies and classifies workplace objects in zenithal (overhead) photos, including bins, cleaning products, food waste, cups, glasses, backpacks and handbags. Based on dynamic rules, the model also categorises each area as clean or messy.
The second phase focused on detecting people. Precision was crucial here, because images containing identifiable individuals must not be processed or stored. We implemented a person detection model that recognises partial views of people, such as a foot or a hand, without capturing recognisable facial features.
- Object detection and classification of workplace items in zenithal photos
- Clean or messy area classification based on dynamic rules
- Person detection that recognises partial views such as a foot or a hand
- Privacy safeguard: images with people are not processed or stored
The outcome
Although few public datasets of zenithal photos exist, we developed a highly precise model that identifies objects in workplace environments and distinguishes clean areas from dirty or messy ones.
Object detection and classification give the client actionable information for timely cleaning, while privacy-focused person detection ensures sensitive information is handled responsibly.
