Are Humans Necessary for LiDAR Point Cloud Feature Extraction?
Throughout the history of mapping, technology advances have threatened to make humans obsolete in a number of key roles. In some cases, technology has succeeded. Most often, however, technology at best has become a valuable assistant, one that reduces or eliminates the need for repetitive manual tasks performed by technicians while still requiring humans to achieve in-depth analysis and understanding.
According to Mike Kitaif, manager of software development for Cardinal Systems LLC, automation has allowed today’s mapping technicians to produce the same amount of data today that would have required four technicians in the 1980s. Feature extraction from photogrammetry is now a highly technical, efficient and refined process. The promise has long been that new technology would make the process completely automatic; however, that goal remains elusive. “Automation has had a dramatic impact, but total automation is far from being realized,” he said during a presentation at the recent MAPPS Winter Conference in Miami.
“Are we too quick to underestimate human interaction?” Kitaif asked.
He noted that today’s point cloud industry has similar expectations for automation. Indeed, software has come a long way in the last several years. Modern applications offer the ability to streamline a number of operations in the feature extraction process. One example is cursor draping (a real-time mode in Cardinal Systems’ Vr Mapping software), which eliminates the need for operators to place markers on a ground or feature surface when extracting features. High/low point searching makes it easier to collect curb lines and find low lines even when vegetation is present. Automated classification has improved, and the extraction of power lines from LiDAR point data has become a highly successful automated process.
However, challenges remain. Uneven road edges, broken concrete, overgrown vegetation, shadows and sparse point distribution in point cloud data make complete automation difficult to attain. “Human interpretation continues to play an important role in producing intelligent data, using semi-automated feature extraction as a helper,” Kitaif said.
He believes feature extraction from point cloud data will follow the same path as photogrammetry. However, the advances are occurring faster and more dramatically. “Because of the complexity of the features we are attempting to collect, automated feature extraction is still a goal for the future,” he said, noting that software developers will continue to push the technology.
“Each of the software development efforts might have a very different approach to solving the puzzle of feature extraction challenges,” he added, “but advances will be made, and cost-effective solutions will be the result.”