Instance Segmentation of Images Above the Ceiling Using Mask R-CNN

Above the ceiling is a space between a suspended ceiling and a true ceiling of each floor in office buildings. It is used for installing heating, ventilation, and air conditioning (HVAC) equipment, electrical conduit, and more. Inspecting this space needs to be proceeded with caution due to its fragile tiles, low height, darkness, and multiple obstructions. Automatic retrieving and understanding of information of this space has the HVAC industry's attention since it helps reduce work on inspecting and maintaining the current state of the equipment. To explore to what extent a machine can understand objects in the above ceiling space, 701 images are collected and annotated. Six groups of prevalent objects are trained with the instance segmentation Mask R-CNN architecture. The most prevalent objects found above the ceiling, namely, threaded rod and cable, can be detected at F1 scores of 0.96 and 0.88, respectively. Metal bar and pipe, which are found to be approximately half of the most prevalent objects, have F1 scores of 0.89 and 0.79, respectively. The least found common objects, namely, duct and HVAC-E, have F1 scores of 0.69 and 0.82, respectively.