Object detection methods on compressed domain videos: An overview, comparative analysis, and new directions

A review of the object detection on compressed domain video is presented in this paper. The literature reviewed spans over the past 32 years (1990-2022), covering the key methods for object detection in video compression. The comprehensive overview of the state-of-art methods is conducted under the MPEG-2, H.264, and HEVC compression standards. We highlight two taxonomies: the way of using motion vector information in the techniques and the video compression standards involved for object detection. In the taxonomy for motion vector information, the object detection methods of using motion vector alone, without using a motion vector, and a mixture of using motion vector and other indexes are deeply discussed. In addition, we provide a unified framework for analyzing the object detection methods in terms of various mainstream video compression standards. Furthermore, the challenges of object detection in compressed domain videos are summarized. Finally, we provide some guidelines for object detection on compressed domain videos and identify a range of future research opportunities.