This case study is based on exploratory work carried out on the Sošice construction site.
Experience the site in the video format:
Improving sustainability in construction is not just about using greener materials or ticking certification boxes. A big part of the problem often sits much closer to the site: we simply do not see clearly enough what is actually being built, when it is being built, and how fast materials are being used. When that visibility is missing, over-ordering, inefficient work sequences, and unnecessary waste quickly follow.
The goal of this case study was to understand whether multimodal AI could help improve early-stage sustainability efforts by making construction activities, material placement, and productivity patterns more visible during day-to-day site operations.
Case Context and Monitoring Strategy
The Sošice site was used as the main testing environment. Instead of relying on a single data source, the monitoring approach combined several types of visual information to capture progress from different angles.
The following data sources were used:
- Drone videos to track overall site development and exterior progress
- Handheld action cameras to record continuous interior work activities
- Site photos and short video clips for close-up observation of specific tasks
- LIDAR scans as a reference for checking spatial accuracy
Fig 1. Case study project floor plan
Each source played a different role. Drone footage helped with the big picture, showing how the site evolved over time. Action cameras and photos made it possible to observe individual work cycles and how materials were handled on site. LIDAR data served as a control reference, allowing visual estimates to be checked against actual geometry.
All collected material was reviewed to reconstruct construction sequences and spot areas where better monitoring could reduce uncertainty around material use and productivity, both of which directly affect waste.
Quantity Monitoring as a Proxy for Waste Detection
 After data capture, the focus shifted to quantity monitoring, with masonry work selected as a test case due to its high material volume and waste potential.
The aim was not to achieve contract-level measurement accuracy. Instead, the question was simpler: can rough, AI-based quantity estimates already tell us something useful from a sustainability perspective?
The workflow followed these steps:
- Extracting frames from recorded videos
- Cropping images to focus on active work zones
- Uploading images into Google AI Studio
- Using task-specific natural-language prompts
- Refining prompts to improve consistency
- Recording brick counts at the start and end of activities
- Converting counts into wall volume using standard brick dimensions
- Comparing results with manually recorded data
Fig 2. AI-based brick counting
This process was intentionally lightweight. Rather than chasing perfect numbers, it explored whether approximate quantity tracking could already highlight trends, inefficiencies, or early signs of material waste during construction.
(...)
