Fig. 4
Example of how results from four different video replicates from two different dives of the same transect are automatically cropped to the same area. From top to bottom: the area covered by the orthoprojection of the 3D point cloud can differ depending on how the video was taken (top row). By using the transect line of the semantic orthoprojection (second row) with known length to give a scale (50m in this example), the RGB and semantic orthoprojection can be cropped to a metric distance (here 1m) around the transect line for increased consistency of surveying results (third and fourth row).
