CVVisual: Interactive visual debugging of computer vision programs

Abstract

Most computer vision applications are built from a combination of basic computer vision algorithms, such as filters, descriptors and matchers. The functionality of this computer vision toolbox is well understood and solid implementations exist. One of the leading and most often used implementations is the Open Source Computer Vision Library (OpenCV), which implements more than 500 computer vision algorithms. However, most of these algorithms have multiple parameters that have to be tuned to the specific vision application. Therefore, during development of new vision applications, some human-in-the-loop iteration cannot be avoided. The problem is that OpenCV lacks support for this phase of development. This usually leads to a situation where each developer creates his own static ad-hoc debug solutions that output intermediate results in a printf-debugging manner. To remedy this situation, we developed reusable general-purpose interactive tools for visual debugging and development of OpenCV-based computer vision applications.