Object detection for building automation schematics

Abstract

In Building Automation projects, a large part of time is spent on reading schematic information that describes how the final system should work. This is often a laborious and time-consuming process. Additionally, this process is as well fault prone, as reading the same style documents for a long period can cause fatigue. RAU-Service, A Western-Northern Finnish based Building Automation solution provider is trying to solve this issue by using computer vision-based algorithms to automatically read schematic information.

To solve the problem of providing such a solution to an entire corporation, a soft-ware-as-a-service architecture is used. This thesis describes the process how an object detection algorithm was designed to detect important information in various Building Automation schematics. By using computer vision-based algorithms, it can detect key features in schematics to determine the precise layout. Using template matching to detect individual symbols, the device information of the schematics is obtained. Finally, the textual information is obtained using the Tesseract OCR model on a filtered schematic page, acquiring important textual data for the previously identified devices.

Several schematics from two different companies were tested and mean average precision (F1 score) across all categories was 92%. Lastly, the thesis delves into future development of this system and its potential use in production.