Automatic far‐field camera calibration for construction scene analysis

The use of cameras for safety monitoring, progress tracking, and site security has grown significantly on construction civil infrastructure sites over the past decade. Localization resources is a crucial prerequisite many applications in automated management. However, most existing vision-based methods perform analysis image plane, overlooking effect perspective depth. manual labor-intensive process traditional calibration techniques, as well busy restrictive environment, makes this challenging task. This study proposes framework automatic camera with no intervention. utilizes convolutional neural networks geometrical scene object detection, which are used to estimate location horizon line, vertical vanishing point, objects known height distributions. enables estimation parameters retrieval scale. proposed evaluated images from two major projects Melbourne, Australia. Results show that method achieves minimum accuracy 90% estimating proximity points ground can facilitate further development solutions productivity analysis.