Video Person Re-Identification using True-Color and Grayscale Images

Abstract

Re-identifying a missing person is a crucial step in many forensics investigations. The vast majority of currently
available techniques for re-establishing the identities of missing by the use of several color-accurate cameras.
Due to camera failure or specific processing for gray mode, acquired pedestrian footage may sometimes be in
grayscale in practice.
Person re-identification from true-color to grayscale pedestrian recordings, which we refer to as color to gray
video person re-identification (CGVPR), is required in such scenarios. However, the CGVPR issue is very
difficult because to the fact that the color information that is highly crucial to depict a pedestrian is often
intensity information and monochromatic in grayscale movies. We present a Semi-coupled Dictionary Pair
Learning (SDPL) method based on asymmetric within-video projection to ease the pain points of CGVPR.
SDPL learns a semi-coupled mapping matrix in addition to a true-color and grayscale dictionary for use inside
videos at the same time. The within-video projection matrices you've learned can reduce the file size of any
video, whether it's in color or black and white. The attributes of full-color and grayscale films may be reconciled
with the aid of the learned dictionary pair and the mapping matrix. We create CGVID (color and grayscale video
person reidentification dataset), the first dataset of its kind dedicated to pedestrians. Each of the more than fifty
thousand frames in our collection was captured in a genuine environment. Extensive assessments show that the
gathered CGVID dataset is quite difficult, and it may be utilized for future study of person re-identification.
Evidence from experiments demonstrates
Xiao-Yuan Jing (email: jingxy 2000@126.com) and Zhiping Peng (email: pengzp@foxmail.com) are the writers
who may be reached through email.
F. Ma may be reached at mafei0603@163.com. He is affiliated with the Computer Schools at Guangdong
University of Petrochemical Technology in Maoming, China; Wuhan University in Wuhan, China; and
Pingdingshan University in Pingdingshan, China.
X. Zhu works at the Henan Key Laboratory of Big Data Analysis and Processing and the School of Computer
and Information Engineering at Henan University in Kaifeng, 475001, China. His name appears among the
others as a co-author.
Z. Tang may be reached at tang.zm@mail.njust.edu.cn or via the School of Computer Science and Engineering
at Nanjing University of Science and Technology in Nanjing 210094, China.
Z. Peng may be contacted at pengzp@foxmail.com, and he works in the School of Computer at the Guangdong
University of Petrochemical Technology in Maoming, China.