Demolition Guide

Artificial Intelligence helps selective demolition

Sloopwijzer does automatic material recognition in building facades using Artificial Intelligence (AI). By unleashing AI on photos of buildings, the system estimates the materials used in the façade and whether they provide more value in selective demolition. In time, Sloopwijzer could help draw up demolition inventories and motivate companies and individuals to demolish selectively.

The Sloopwijzer was developed by VITO in close cooperation with Immoterrae and stakeholders VCB, Tracimat, FLOOW2, BOPRO and the City of Leuven.

Specifically, this project is a proof of concept focusing on two different analyses. On the one hand, we used deep learning to detect windows in the facades of buildings in Leuven. On the other hand, some building typologies were created that provide information about the material composition of a building, based on visual aspects. We also tested this detection method on Leuven's building heritage.

Furthermore, we used new data sources (through web scrapping) to estimate the possible residual value of a certain building material, namely used windows.

Finally, we mapped out the possible innovation pathways that could result from this project, and which stakeholders in the building sector would benefit.

MOST IMPORTANT
RESULTS

  1. The window detection model, adopted from the international literature, has an average accuracy of 85% for an image. Some building styles perform remarkably less; refining the algorithm with training data for typical building styles in Flanders can improve this.
  2. We applied three AI models for building typology recognition to our own dataset of about a hundred facades from Flemish cities. The accuracy is above 90% in each case. The brick detection model even achieved 95%.
  3. We performed a web scraping that collected data from websites for second-hand building materials (windows). This way, we found out which properties influence the residual value the most: surface area, type of glass, window profile, and window type.
  4. We discussed the analyses with our partners and the various stakeholders. Thanks to the network of this partnership, we got a good insight into which market needs our technology can address.

 

MOST IMPORTANT
LESSONS LEARNED

  1. Google Streetview as a publicly available data source has many advantages, but also limitations. The systems developed in this project are a good starting point to do material balance estimates at the district or city level that are much more accurate than currently available analyses.
  2. The possibilities of AI detection are very numerous. The more specific the demand or need for this type of analysis, the better a workflow can be developed that can answer the question.
  3. Circular economy is difficult to monitor. For instance, determining material residual values is a major challenge for which there is no ready-made methodology yet. Therefore, we looked for alternative data streams, such as via web scrapping. It was innovative and inspiring to work in this way.
  4. The construction industry is a complex sector with many different parties. Under the impetus of circular economy, roles such as urban miners, reuse initiatives, data platforms ... will be added. To change the current linear value chain, you need to engage with many different parties.
2 AI detection systems applied
3 AI models for building typology recognition
85% up to 95% accuracy

WHAT DOES
THE FUTURE HOLD?

During the project, we investigated possible follow-up paths, such as focusing on a particular material, the differences in quality or extending the project to a wider region. Through conversations with stakeholders in the construction industry, we also gauged their specific interest, needs and preferences in this trajectory. Based on this input, preparations for several follow-up projects were started, with the aim of further scaling up both AI detection for materials and residual value estimation.