A multi-tool for ground-truth stereo correspondence, object outlining and points-of-interest selection

Abstract

In this paper we present architecture and functionality of the visual multi-tool designed for acquisition of ground-truth and reference data for computer vision experimentation. The multi-tool allows three main functions, namely manual matching of the corresponding points for multi-view correlation, outlining of the image objects with polygons, as well as selection of characteristic points in specific image areas. These functions allows gathering of experimental data which are used for training and/or verification in such computer vision methods as stereo correlation, road signs detection and recognition, as well as color based segmentation. We present overview of the experimental results which were made possible with this multi-tool, as well as we discuss its potential further applications and extensions. The presented software platform was made available on the Internet.