An Improved Evaluation Method for Optical Flow of Endpoint Error

DOI

10.2991/cnct-16.2017.44How to use a DOI?

Keywords

Optical flow, Evaluation, Benchmark, Corrected

Abstract

A corrected evaluation benchmark is proposed based on the Middlebury benchmark. The major contributions of the presented benchmark include the following: error metrics of endpoint error are modified, and the corrected metrics can decrease the detrimental influence of human factors and reflect the actual flow evaluation performance. In addition, the metric of normalization endpoint error is projected to reveal the relative error of the flow result endpoint, which can eliminate the influence of small flow errors. For flow error statistics, a mathematical expectation is employed instead of the current mathematical average as the expectation can reflect a more appropriate error distribution for the flow result. In addition, robust statistics and the accuracy measures derived from the Middlebury benchmark are employed to evaluate the robustness of optical flow in the proposed method.

Copyright

© 2017, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Cong-xuan ZHANG
AU  - Ling-ling ZHU
AU  - Zhen CHEN
AU  - Ding-ding KONG
AU  - Xuan SHANG
PY  - 2016/12
DA  - 2016/12
TI  - An Improved Evaluation Method for Optical Flow of Endpoint Error
BT  - Proceedings of the International Conference on Computer Networks and Communication Technology (CNCT 2016)
PB  - Atlantis Press
SP  - 312
EP  - 317
SN  - 2352-538X
UR  - https://doi.org/10.2991/cnct-16.2017.44
DO  - 10.2991/cnct-16.2017.44
ID  - ZHANG2016/12
ER  -