Change detection in drone-captured image data for the construction sector: Exploring the possibilities and obstacles of implementing automatic progress monitoring in a dynamic industry

Abstract

Low productivity, expected labour shortages and hazardous work environment are some of the construction sector’s problems. Digitalisation has revolutionised many other industries, and the construction industry is starting to realise the possibilities this technology can have to meet the industry’s challenges. However, the fragmented and project-based construction industry and the dynamic project processes, that characterise construction projects, present many obstacles to effective progress monitoring. Today’s monitoring progress is primarily manual, subjective, and irregular. This, in turn, leads to late changes, errors and delays, and often cause construction projects to fail to meet deadlines and budgets. Using change detection algorithms to identify changes that have happened over time in drone-captured images could facilitate the work on site by, for example, quickly identify areas of interest, give early warnings of deviations from the design documents, and highlight safety concerns. Therefore, the aim with this master thesis is to investigate how AI can be utilised to automatically monitor construction project progress from drone-captured data. Also, the characteristics of the industry that can affect the use of such a system are studied. The method of this thesis consisted of a literature review, pre-study interviews, a study visit and code testing. The results indicate that changes can be detected using AI on drone-captured data. However, adjustments or improvements need to be made for this to be truly useful. The results show multiple areas where this type of process needs to be adjusted to improve accuracy and make sure this method fits automatic progress monitoring of the dynamic construction industry. There exists, a lack of datasets and national model libraries on construction objects and images, as well as a lack of advanced digital knowledge and competences in the workforce. Nevertheless, change detection in images captured by drones could be used to address challenges such as safety, productivity and labour shortages. However, this will require a rigorous routine that describes how to collect, analyse, store and handle the data as well as frequency.