Breaking the Boundaries: Integrating GIS, AI, and Lidar for Digital Innovation

A key component of the digital transformation of GIS data is the rapid integration and interrogation of large data sets, lidar point clouds, and visual information to support rapid decision making. For example, smart city automation relies on accurate knowledge of street infrastructure, such as the location and condition of utility poles and power lines, while forestry managers require canopy and trunk estimation and digital terrain mapping to enhance their operations. Significant challenges remain, including the need for rapid but accurate processing of spatial point clouds and the integration of visual and multispectral data with lidar data to provide information beyond a simple (x,y,z) location.

Traditional feature-based methods for processing point cloud data can be slow and imprecise. In some cases, such as on mine sites, feature-based algorithms may misinterpret ground profiles as buildings due to the planar shapes of spoil heaps, leading to inaccuracies. In contrast, at Spacesium we are using cutting-edge deep learning algorithms, such as R-CNN, DeepLab v3 and PointNet++, in conjunction with semi-supervised retraining, to rapidly segment and classify point clouds utilizing cloud scale resources.

The integration of visual and multispectral data presents a unique set of challenges and opportunities. While it allows for more than just the specification of the position of an xyz point cloud, using both location and visual data can enable the determination of viable routes for autonomous vehicles. On the challenge side, it is computationally expensive to fuse these data sets that are captured in separate imaging systems. At Spacesium, we have implemented custom algorithms based on tools from Image Processing Toolbox™, Computer Vision Toolbox™, and Lidar Toolbox™ to efficiently correct, compute, and fuse these data sets.