Selective frame analysis for efficient object tracking: balancing speed with accuracy in MOT systems

Yubraj Jung Shah, Yi Guo, Laurence A. F. Park, Oliver Obst

Research output: Chapter in Book / Conference PaperChapterpeer-review

Abstract

Applications such as autonomous driving and video surveillance rely on Multiple Object Tracking (MOT) technology to accurately identify objects in video data. Real-time MOT systems are often challenged for continuously improving computational efficiency while maintaining the acceptable level of accuracy. For every frame, advanced algorithms for detection and tracking is used to identify and track objects. However, it is still unclear if they should be used across all frames, or whether it would be better to use the algorithm only for selected frames. This is an empirical question, best answered by experimental research. Here, we explore how frame skipping during object detection impacts tracking accuracy and speed in real-time MOT systems. We examined the trade-off between skipping and tracking robustness for given MOT tasks. The consequences of frame skipping were evaluated using publicly available MOT datasets (KITTI, MOT16, MOT17 and MOT20) in different skipping image frequencies. Frame skipping allowed us to achieve a negligible drop in the MOTA and HOTA score while giving us a big 80% boost in speed over regular baseline configuration.

Original languageEnglish
Title of host publicationMulti-disciplinary Trends in Artificial Intelligence: 17th International Conference, MIWAI 2024, Pattaya, Thailand, November 11-15, 2024, Proceedings, Part I
EditorsChattrakul Sombattheera, Paul Weng, Jun Pang
Place of PublicationSingapore
PublisherSpringer
Pages245-256
Number of pages12
ISBN (Electronic)9789819606924
ISBN (Print)9789819606917
DOIs
Publication statusPublished - 2025

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume15431
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Keywords

  • Computational Efficiency
  • Frame Skipping
  • Multiple Object Tracking (MOT)
  • Selective Frame Skipping
  • Tracking-by-Detection

Fingerprint

Dive into the research topics of 'Selective frame analysis for efficient object tracking: balancing speed with accuracy in MOT systems'. Together they form a unique fingerprint.

Cite this