Object recognition in ground plans

Dieser Inhalt wurde auch auf der internationalen Konferenz „AI in AEC“ am 24.03.2021 von Patrick Hemmer (KSRI-KIT) präsentiert.

In recent years, more and more companies in the architecture, engineering and construction industries have become aware of the potential of artificial intelligence (AI) to improve work processes. This trend is being driven by various technological advances, such as the increasing adoption of Building Information Modeling (BIM). To date, however, few medium and large companies have been able to use their data profitably. As in many other industries, employees are uncertain about the impact AI will have on their work and are therefore reluctant to support the upcoming transformation.

The concept of "human-centered artificial intelligence" has become increasingly important in recent years, both in practice and in science. The approach is based on the idea that AI systems should not replace employees but complement and empower them. Especially in the field of architecture, where many work processes are characterized by a large number of repetitive tasks, identifying and assigning these tasks to computers is a promising endeavor. For this reason, it can be assumed that more and more systems based on the combination of human and artificial intelligence will find application in practice in the future.

The process of mass determination requires the manual counting of relevant components. In the area of building operations, the challenge here is to identify objects from floor plans, which are usually only available as rasterized images or printouts (see Figure 1). 

Figure 1: Reduction of manual work through AI applications

To solve this problem and promote quality control, the human-in-the-loop system developed as part of the Smart Design and Construction research project recognizes symbols relevant to users in scanned floor plans to support the design process and simplify matching with building requirements (see Figure 2).

Figure 2: By quantifying prediction uncertainty, the system can communicate to users which symbols need to be verified

Here, a bilateral cooperation between user and system takes place. On the one hand, it supports its users by providing relevant recommendations, but on the other hand, it also offers them the opportunity to contribute their domain knowledge in a targeted manner in order to ultimately contribute to more efficient and reliable mass determination.

In addition to this use case, a large number of development teams are working on solutions for other use cases as part of the research project in order to make a decisive contribution to the digitalization of the construction industry.