Industry Analysis from SPAR 2015 & Introduction of Deep Learning

Recently, Civil Maps made it’s debut appearance at SPAR International 2015. It was a great opportunity for us to examine the state of the art in the industry, highlight some of the strengths and weaknesses in processing point cloud data.

Generally speaking, the scanning technology is getting drastically better. The LiDAR technology, along with the cloud registration processes are rapidly advancing. Algorithms that create point clouds are generating vast amounts of 3D data, creating a data bottleneck.

Making maps is an interdisciplinary problem statement. It means that the current workflow for mapping asset information from unorganized 3D data will become the main bottleneck. While some tools attempt to automate feature extraction, they often live on desktop computers and are not application agnostic.

What can we do to solve this problem? Civil Maps is creating an application agnostic approach to map generation by training computers to create maps from basic primitive descriptors.

Traditionally, in image processing feature descriptors are used to annotate 2D intensity and color information. These same principles can be leveraged to create feature descriptors in Euclidean space, color space, and temporal space. Leveraging this makes it possible to create thousands of feature descriptors.