Automated computer vision-based detection of components of under-construction indoor partitions

This paper presents a computer vision-based algorithm that automatically detects the components of an interior partition and infers its current state using 2D digital images. The algorithm relies on four integrated shape and color-based modules, which detect studs, insulation, electrical outlets, and three states for drywall sheets (installed, plastered, and painted). Based on the results of the four modules, images are classified into five states. The proposed method was validated using three image databases of indoor construction sites captured by a quadcopter (a type of unmanned aerial vehicle), a smartphone, and collected from publically available sources on the internet. The method's high accuracy rates, its fast performance, and applicability to different contexts such as automated robotic inspection are indicative of its promising performance. The visual detection results can potentially provide situational awareness for construction trades, provide future progress tracking systems with information on actual state, and help leverage the use of image processing at indoor sites.