Evaluation of Stereo Matching Systems for Real World Applications Using Structured Light for Ground Truth Estimation

In this paper we present an evaluation method for stereo matching systems and sensors especially for real world indoor applications. We estimate ground truth reference images by illuminating scenes with structured light. The paper starts with the selection of appropriate scenes, goes over ground truth estimation to the finally resulting evaluation of the stereo sensors. Three different stereo vision sensors are tested with three different input scenes. Beside the subjective review of the disparity map images a pixel wise evaluation is carried out. We mainly describe an evaluation method for existing stereo sensors which should help developing new ones. 1 Motivation The need of dense 3D information to support robots and autonomous systems is increasing permanently. Service robots require dense depth information to fulfill their tasks [4] and intelligent vehicles rely on accurate sensor information to support navigation along tracks and avoid obstacles [1]. Stereo vision is a dedica...

• Type: Conference
• Year: 2007
• Where: MVA
• Authors: Martin Humenberger, Daniel Hartermann, Wilfried Kubinger
• Keywords: Computer Vision, Disparity Maps, MVA 2007, Stereo, Stereo Vision