Differential Features for Pedestrian Detection: A Taylor Series Perspective

Differential features are popularly used in computer vision tasks, such as object detection. In this paper, the authors revisit these features from a functional approximation perspective. In particular, the authors view an image as a 2-D functional and investigate its Taylor series approximation. Differential features are derived from the approximation coefficients and, therefore, are naturally collected for appearance representation. Thus motivated, the authors propose to use the zeroth-, first-, and second-order differential features for pedestrian detection and call such features Taylor feature transform (TAFT). In practice, the TAFT features are computed by discrete sampling to address scale issues and meanwhile achieve computational efficiency. In addition, orientation insensitivity is handled by using directional versions of differentials. When applied to pedestrian detection, the TAFT is sampled on grid pixels and calculated from multiple channels following previous solutions. In the author extensive experiments on the INRIA, Caltech, TUD-Brussel, and KITTI data sets, the TAFT achieves state-of-the-art results. It outperforms all handcrafted features and performs on par with many deep-learning solutions. Moreover, when a low false-positive rate is requested, the TAFT generates results that are better than or comparable to the state-of-the-art deep learning-based methods. Meanwhile, the author implementation runs at 33 fps for 640×480 images without GPU, making TAFT favorable in many practical scenarios.

• Record URL:
  • https://doi.org/10.1109/TITS.2018.2869087
• Availability:
  • Find a library where document is available. Order URL: http://worldcat.org/oclc/41297384
• Supplemental Notes:
  • Copyright © 2019, IEEE.
• Authors:
  • Shen, Jifeng
  • Zuo, Xin
  • Yang, Wankou
  • Prokhorov, Danil
  • Mei, Xue
  • Ling, Haibin
• Publication Date: 2019-8

Language

• English

Media Info

• Media Type: Digital/other
• Features: Figures; References; Tables;
• Pagination: pp 2913-2922
• Serial:
  • IEEE Transactions on Intelligent Transportation Systems
  • Volume: 20
  • Issue Number: 8
  • Publisher: Institute of Electrical and Electronics Engineers (IEEE)
  • ISSN: 1524-9050
  • Serial URL: http://ieeexplore.ieee.org/xpl/RecentIssue.jsp?punumber=6979

Subject/Index Terms

• TRT Terms: Computer vision; Data files; Experiments; Image analysis; Pedestrian detectors; Pedestrians; Visual texture recognition
• Subject Areas: Data and Information Technology; Highways; Pedestrians and Bicyclists;

Filing Info

• Accession Number: 01715797
• Record Type: Publication
• Files: TLIB, TRIS
• Created Date: Sep 3 2019 9:15AM