Analysis of Efficiency and Accuracy of Plane Fitting Methods to Enable Plane Detection Technology

Aaron Turrill

Abstract


This report investigates the usage of least squares and vectors to find the dominant plane in point cloud data. New sensor solutions such as the Occipital Structure Sensor permit a more simple, lightweight solution to find 3 dimensional representations of surfaces. This lightweight solution could be used with existing Unmanned Autonomous Systems (UAS), especially smaller airborne systems that have smaller limits on their payload; when coupled with less powerful, smaller computing systems, these sensors may permit autonomous landing. In this project, testing on methods of dominant plane fitting for point cloud data was conducted using vectors and least squares methods to determine which is most efficient and which is most accurate. A solution to cluttered environments is also presented in the form of dividing the point cloud data returned into smaller samples and fitting planes to each sample. Testing of these methods found that vectors are on average faster than the least squares method at the expense of accuracy, however these time efficiencies are reduced as the least squares method is subsampled.

Full Text:

PDF