The optical flow, the disparity, and the scene flow variables are estimated by minimizing variational formulations involving a data and a smoothness term. Both of these terms are based on assumptions of gray value consistency and smoothness which may not be exactly fulfilled. Moreover, the computed minimizers will generally not be globally optimal solutions. For follow-on calculations (e.g. speed, accuracy of world flow, detection of moving objects, segmentation of objects, integration, etc.), it is therefore of utmost importance to also estimate some kind of confidence measure associated with optical flow, disparity and scene flow. In this chapter, the error characteristics for respective variables are analyzed and variance measures are derived from the input images and the estimated variables themselves. Subsequently, scene flow metrics are derived for the likelihood of movement and for the velocity of a scene flow vector.
Everything should be as simple as it is, but not simpler. Albert Einstein
Hirschmüller, H.: Stereo vision in structured environments by consistent semi-global matching. In: Proc. International Conference on Computer Vision and Pattern Recognition, New York, NY, USA, pp. 2386–2393 (2006)
Mahalanobis, P.C.: On the generalised distance in statistics. In: Proc. of the National Institute of Science of India, India, vol. 12, pp. 49–55 (1936)
Shimizu, M., Okutomi, M.: Precise sub-pixel estimation on area-based matching. In: Proc. International Conference on Computer Vision, Vancouver, Canada, pp. 90–97 (2001)
Vaudrey, T., Rabe, C., Klette, R., Milburn, J.: Differences between stereo and motion behaviour on synthetic andreal-world stereo sequences. In: Online-Proc. Image and Vision Computing New Zealand, Christchurch, NZ, November 2008
Wedel, A., Cremers, D. (2011). Motion Metrics for Scene Flow.
In: Stereo Scene Flow for 3D Motion Analysis. Springer, London. https://doi.org/10.1007/978-0-85729-965-9_5
