Fast Object Detection in HEVC Intra Compressed Domain

Liuhong Chen, Heming Sun*, Jiro Katto, Xiaoyang Zeng, Yibo Fan

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contribution

3 Citations (Scopus)


Conventional object detection methods are in the pixel domain and require full decoding with high computational complexity. In this paper, we propose a fast object detection method in the intra compressed domain of High Efficiency Video Coding (HEVC), which significantly accelerates the object detection process that uses compressed video images. Considering the characteristics of various coding features, we select 3 types of data for object detection, including partitioning depths, prediction modes, and residuals. To achieve a more discriminative representation of the residuals, we design an iterative restoration algorithm that can generate the details of the original image and reduce the noise in the residuals. Extensive evaluations on multiple HEVC test sequences and large-scale object detection dataset BDD100K confirm the effectiveness of our method. With a slight reduction in detection accuracy, our compressed domain detection system runs 1.8 times faster than the pixel domain.

Original languageEnglish
Title of host publication29th European Signal Processing Conference, EUSIPCO 2021 - Proceedings
PublisherEuropean Signal Processing Conference, EUSIPCO
Number of pages5
ISBN (Electronic)9789082797060
Publication statusPublished - 2021
Event29th European Signal Processing Conference, EUSIPCO 2021 - Dublin, Ireland
Duration: 2021 Aug 232021 Aug 27

Publication series

NameEuropean Signal Processing Conference
ISSN (Print)2219-5491


Conference29th European Signal Processing Conference, EUSIPCO 2021


  • Compressed domain video analysis
  • HEVC
  • Intra frame
  • Object detection

ASJC Scopus subject areas

  • Signal Processing
  • Electrical and Electronic Engineering


Dive into the research topics of 'Fast Object Detection in HEVC Intra Compressed Domain'. Together they form a unique fingerprint.

Cite this