Many of our customers and partners are concerned with the accuracy of our automated point-cloud feature extraction method. LiDAR instruments provide excellent representations of real environments, and it would be counterproductive to lose accuracy in the data processing stages. In this post, we quickly examine the difference between the CivilMaps approach and the current industry practices of manual annotation.
With manual annotation, a person uses desktop tools to load the point-cloud from disk, visualize it, and traverse it. In contrast, we at CivilMaps use an Artificial Intelligence engine capable of identifying railroad tracks in a point-cloud, and extracting the centerlines using cloud computing.
Comparison Study
To study our accuracy, we used a dataset of a railroad stretch of about 3 km. The next figure is a “view from above” of some point-cloud data with three plots.
Next, we look at elevation over horizontal distance. Looking at the results, we can see that the GIS consultant who performed the manual annotation created a segmented line which is choppy and irregular. In contrast, the CivilMaps output is much smoother and a more accurate representation of the real infrastructure. We observed errors up to 2 cm, in this particular experiment.