Multi-View Matching for Onsite Construction Resources with Combinatorial Optimization

When multiple video cameras are set up to monitor construction activities, onsite construction resources (e.g. equipment and workers) might be captured by two or more cameras at the same time. It becomes important to identify whether the resources captured into separate camera views refer to the same one on the site. Otherwise, the automatic reporting of onsite resources utilization will produce repetitive countings. This paper proposes a novel method for matching onsite construction resources in multiple camera views. The method relies on the visual features of a construction site and the spatial relationships of the resources on the site as matching cues. It starts with searching potential matching candidates between the camera views following their epipolar constraints. Then, the candidates’ local triangular coordinates are calculated to define matching costs. This way, the matching of multiple construction resources between camera views could be solved through combinatorial optimization. The proposed method has been tested to match workers, equipment and traffic cones within the images and videos captured from construction sites. The test results showed that the method could reach an average of 93% matching accuracy.