Performance Evaluation of Stereo and Motion Analysis on Rectified Image Sequences

This paper introduces into seven real-world road driving stereo sequences (provided by Daimler AG, Germany; now freely available for academic research); it also informs about their use for performance evaluation in some experiments using common stereo and motion analysis algorithms. Often, such algorithms are tested on a few frames only, or on synthetic sequences, but not on long real-world sequences. The provided sequences have 250 to 300 stereo pairs each; they have been used at Daimler AG for testing 6D vision, which fuses disparity and motion data (by using a Kalman filter). In this paper we introduce into those seven sequences (and inform about the download web site); we discuss a few approaches how to use those sequences for testing either stereo or motion algorithms, or both combined together. Certainly, those sequences do have many more potentials for future performance evaluations for stereo and motion analysis algorithms.