New visualization techniques could help create 3D models that are easier to use

Digital 3D models have helped rescuers analyze hazardous environments, scientists monitor natural resources and autonomous cars navigate city streets, but the large amount of data needed to create the models requires specialized software and expertise, boosting costs and limiting more widespread adoption.

That could change with the help of research led by Federico Iuricich, an assistant professor in Clemson University's School of Computing.

For the project, Iuricich is focusing on 3D point clouds, the basic building blocks of 3D models.

A 3D point cloud is created when a sensor, such as a LiDAR, shoots light at an object and measures the time it takes to bounce back from the surface. The resulting collection of sample points, also called a scatter plot, represents the scanned object in three dimensions.

Analyzing 3D point clouds is still a time-consuming and very labor-intensive operation, Iuricich said. This problem is a result of looking at point clouds only as 3D scatter plots, or as realistic reconstructions of the three-dimensional world.

While achieving realism is essential for creating digital twins of the real world, it can be detrimental when supporting specific human tasks, such as helping environmental scientists identify a single tree in a forest or quickly determining which parts of a river are most susceptible to flooding, he said.

Iuricich instead envisions a future in which 3D point clouds will be analyzed by humans as easily and efficiently as it is now to look at images or watch videos.

He is laying the groundwork with his new research, which is funded by a CAREER award from the National Science Foundation.

As part of the project, he and his group plan to investigate and develop new point-cloud visualization techniques that would be assisted by artificial intelligence. The idea is to develop techniques that can efficiently support point cloud visualization by removing unnecessary information, similar to how road maps avoid showing every last detail during navigation.

Iuricich will then work with collaborators in Clemson's departments of Forestry and Automotive Engineering to test the techniques in real-world applications.

"The problem at the moment is that there is no empirical evidence indicating what type of visualization could support these tasks," Iuricich said. "Granted that 3D scatterplots are not helping, we want to generate metrics and rankings to establish precisely what visualization technique will allow you to, for example, count trees much faster or give you better situational awareness while collaborating with an autonomous vehicle."

Iuricich is also creating educational programs based on his point-cloud research. For example, one program for high school students will turn his visualization techniques into an educational game aimed at identifying trees in the Clemson Experimental Forest.

Iuricich also plans to work with PEER & WISE to recruit undergraduate students and mentor them through independent projects.

"I am excited because the CAREER award allows me to not only work on the research problem for five years but also to open opportunities for the educational component as well," he said.