Open AccessFast background estimation on long video sequence
Author(s) -
Fu HuiNi,
Wang BenZhang,
Liu HengZhu
Publication year - 2019
Publication title -
electronics letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.375
H-Index - 146
ISSN - 1350-911X
DOI - 10.1049/el.2019.1178
Subject(s) - robust principal component analysis , computer science , frame (networking) , set (abstract data type) , artificial intelligence , sequence (biology) , computer vision , matrix completion , principal component analysis , matrix (chemical analysis) , image (mathematics) , pattern recognition (psychology) , algorithm , telecommunications , physics , materials science , quantum mechanics , biology , gaussian , composite material , genetics , programming language
Background estimation is essential in many computer vision applications with video frames. It refers to two important issues: accuracy and efficiency. There are mainly two challenges: how to recover the background image from as few frames as possible, and how to get a background image as fast as possible from a long video sequence. This Letter proposed a fast background estimation approach based on matrix completion to solve the second issue. Considering that matrix completions set all frames to be column vectors in a big matrix. While sequential images are to some extent related and redundant frames exist, the authors’ approach implements a frame selection step to decrease the number of frames before using fast robust principal component analysis (RPCA) to deal with matrix completion problem. Their experiments selected long video sequences from scene background initialisation (SBI) dataset. Results show that compared with existing RPCA matrix completion algorithms and other state‐of‐the‐art methods, their method is much better in processing efficiency while still keeps the same good performance.