End-to-end learning of local point cloud feature descriptors

Thumbnail Image
Wehr, David
Major Professor
Rafael Radkowski
Yan-Bin Jia
Committee Member
Journal Title
Journal ISSN
Volume Title
Research Projects
Organizational Units
Organizational Unit
Computer Science

Computer Science—the theory, representation, processing, communication and use of information—is fundamentally transforming every aspect of human endeavor. The Department of Computer Science at Iowa State University advances computational and information sciences through; 1. educational and research programs within and beyond the university; 2. active engagement to help define national and international research, and 3. educational agendas, and sustained commitment to graduating leaders for academia, industry and government.

The Computer Science Department was officially established in 1969, with Robert Stewart serving as the founding Department Chair. Faculty were composed of joint appointments with Mathematics, Statistics, and Electrical Engineering. In 1969, the building which now houses the Computer Science department, then simply called the Computer Science building, was completed. Later it was named Atanasoff Hall. Throughout the 1980s to present, the department expanded and developed its teaching and research agendas to cover many areas of computing.

Dates of Existence

Related Units

Journal Issue
Is Version Of

Emerging technologies like augmented reality and autonomous vehicles have resulted in a growing need to identify and track objects in the environment. Object tracking and localization is frequently accomplished through the use of local feature descriptors, either in 2D or 3D. However, state-of-the-art feature descriptors often suffer from incorrect matches, which affects tracking and localization accuracy. More robust 3D feature descriptors would make these applications more accurate, reliable, and safe. This research studies the use of a pointwise convolutional neural network for the task of creating local 3D feature descriptors on point clouds. A network to produce feature descriptors and keypoint scores is designed, and a loss function and training method is developed. The resulting learned descriptors are evaluated on four different objects, using synthetic and scanned point clouds. The evaluation shows that the descriptors can effectively register objects with noise, and that the keypoint scores can reduce the number of required iterations for registration by a factor of three. An analysis of the learned filters provides insights into what the descriptors encode and potential avenues for improvement.

Subject Categories
Thu Aug 01 00:00:00 UTC 2019