Building Element Recognition with Thermal-Mapped Point Clouds

Automated recognition of building elements convey vital information for inspection, monitoring and maintenance operations in indoor environments. However, existing object recognition methods from point clouds suffer from problems due to sensor noise, occlusion and clutter, which are prevalent in indoor environments. This paper proposes an object recognition method based on thermal-mapped point clouds for building elements consisting of electrical systems and heating, ventilation, and air-conditioning (HVAC) components. The proposed processing pipeline involves data collection from a mobile robot using both laser scanners and a thermal camera where temperature mapping can be performed from thermal images to point cloud. Next, the ceiling region containing the building elements of interest is identified and extracted from the point cloud. Segmentation of peak and valley thermal intensity regions is carried out based on absolute and relative temperature threshold values. The identified point cloud clusters can be each associated with a building element and localized based on the cluster center. The proposed building element recognition method was validated with two sets of laser scan data collected in an indoor laboratory. Experimental results for detection of lighting elements and cooling elements showed that the method achieved an average of 100% precision, 90% recall, and 0.25m root mean squared error (RMSE).