Generation of High-Density Hyperspectral Point Clouds of Crops with Robotic Multi-Camera Planning

Abstract

Hyperspectral imaging and point-cloud based mapping of horticultural crops provide crucial information for precision agriculture applications. In this paper, we present an optimal method of autonomously generating high-density hyperspectral point clouds of objects using a robotic multi-camera suite, without the need for high-cost, large-size hyperspectral cameras. The sensing and imaging fusion are integrated through the use of next-best-view planning for a heterogeneous set of cameras, and a novel information gain metric designed specifically for hyperspectral point clouds. We also introduce a tractable approximation to the IG metric for real-time applications. Both metrics are tested on plant scans with augmented spectral data to demonstrate the capabilities and accuracy of the method.