Egocentric-video-based construction quality supervision (EgoConQS): Application of automatic key activity queries

Abstract

Construction quality supervision is essential for project success and safety. Traditional methods relying on manual inspections and paper records are time-consuming, error-prone, and difficult to verify. In-process construction quality supervision offers a more direct and effective approach. Recent advancements in computer vision and egocentric video analysis present opportunities to enhance these processes. This paper introduces the use of key activity queries on egocentric video data for construction quality supervision. A framework, Egocentric Video-Based Construction Quality Supervision (EgoConQS), is developed using a video self-stitching graph network to identify key activities in egocentric videos. EgoConQS facilitates efficient monitoring and quick review of key activity frames. Empirical evaluation with real-world data demonstrates an average recall of 35.85 % and a mAP score of 6.07 %, highlighting the potential of key activity queries for reliable and convenient quality supervision.