Multi-camera vision-based productivity monitoring of earthmoving operations
Jinwoo Kim, Seokho Chi
Abstract
To be successful in managing earthmoving projects, it is very important to monitor the operational efficiency and productivity of heavy equipment. Researchers have investigated many vision-based methods and demonstrated their high applicability to automated productivity monitoring. However, they primarily focused on developing a single-camera vision-based approach that monitors heavy equipment's movement using video data collected from only one camera, and thus they normally failed in continuous earthmoving productivity monitoring due to its limited visibility; for instance, it would be difficult to understand what dump trucks are actually doing if they disappear from the single-camera's field of view. To address these limitations, this paper proposes a multi-camera vision-based productivity monitoring methodology that analyzes videos captured from multiple non-overlapping cameras at the jobsite. The proposed methodology consists of three main processes: (1) multi-camera placement at different physical locations on site, (2) single-camera vision-based equipment monitoring, and (3) multi-camera vision-based equipment matching (i.e., finding the same object in multiple cameras) for productivity analysis. For validation, the authors conducted experiments using video data of 371,125 image frames recorded from an actual earthmoving site for highway construction, and the results confirmed the potential of precise productivity monitoring with the average 97.6% equipment matching accuracy. To the authors' knowledge, this is the first attempt to monitor the productivity of earthmoving equipment using multiple cameras, and these findings will support to more reliable automated productivity monitoring of earthmoving operations.