Multispectral Data Fusion for Robotic Reconnaissance and Mapping

Abstract

The aim of the paper is to describe the data-fusion from optical sensors for mobile robotics reconnaissance and mapping. Data are acquired by stereo pair of CCD cameras, stereo pair of thermal imagers, and TOF (time-of-flight) camera. The fusion is realized by means of spatial data from a TOF camera to ensure “natural” representation of a robot's environment; thus, the thermal and CCD camera data are comprised in one stereo image presented to a binocular, head-mounted display. The data acquisition is performed using a sensor head, which is placed on an Orpheus-X3 robot; both the head and the robot were developed by our working group. After the geometrical calibration of each sensor, the positions of the sensors in 6DOFs are computed. The corresponding data from the CCD camera and the thermal imager are subsequently determined via homogeneous and perspective transformations. The result consists in an image containing aligned data from the CCD camera and the thermal imager for each eye. TOF camera calibration and its impact to the precision of fusion is described. Although the fusion is used for two different tasks - automatic environment mapping and visual telepresence, the utilised calibration and fusion algorithms are, in principle, the same.