Abstract

The construction industry, being one of the largest industrial sectors in Canada, has been continually searching for automated methods that can be adopted to monitor the productivity, consistency, quality and safety of its construction work. The automated recognition of construction operational resources (equipment, workers, materials etc.) has played a significant role in achieving the full automation in monitoring and control of the construction sites. Considering that construction equipment is one of the main operational resources in executing construction tasks, this research work is focused on automated recognition of such equipment from on-site images. In order to achieve this, it is first necessary to evaluate the construction equipment recognition performances of existing object recognition methods. The currently available object recognition datasets that are used to validate the existing recognition methods contain only limited categories of objects, where construction equipment are not included. As a result, it is unclear whether these methods could be used to recognize construction equipment from on-site images, especially considering that construction sites are typically dirty, disorderly, and cluttered. To fill this gap, this research work proposes to create a standardized dataset of construction equipment images that can be used to measure the construction equipment recognition performances of existing object recognition methods. Almost 2,000 images have been collected and compiled to create the dataset, which covers 5 common classes of construction equipment (excavator, loader, tractor, compactor and backhoe loader). Each image has been annotated with information concerning the equipment class, identity, location, orientation, occlusion, and labeling of equipment components (bucket, stick, boom etc.). The effectiveness of the dataset has been tested on two common object recognition methods in computer vision. The recognition tests imply that the recognition methods can be adopted comprehensively for the recognition of construction equipment with the dataset developed in this research. The performances of these two methods are further compared on the basis of the recognition tests conducted in this work. The results show that the construction equipment recognition performance of existing object recognition methods can be evaluated with the dataset in a standard, unbiased, and extensive way.