Automatic matching of construction onsite resources under camera views

Abstract

When a video camera network is placed on a construction site to monitor onsite activities, construction resources, such as equipment and worker, might be captured by two or more cameras at the same time. Therefore, it is important to conduct the matching to identify whether the resources captured into separate camera views refer to the same one on the site. Otherwise, it leads to the repetitive counting, when analyzing the onsite resources utilization automatically. This paper proposes a novel matching method that relies on the construction site visual features and the spatial relationships of onsite construction resources as the matching cues. Specifically, the method first searches the potential matching candidates between two camera views following their epipolar constraints. Then, the triangular coordinates of these candidates are calculated based on their locations in the triangular mesh of each camera view. This way, the matching of multiple construction resources between two camera views could be converted to a combinatorial optimization problem and solved with the Hungarian algorithm. The proposed method has been tested with the images and videos captured from real construction sites. The test results showed that the average matching accuracy could reach 93%.