Real-time Detection of Activities in Untrimmed Videos

Abstract

The goal of this work is to produce a system for detecting activities in untrimmed videos which runs in real-time with high accuracy on a large number of human and vehicle activities. By real-time we mean that the time between when video files are provided to the system until the time that all detected activities are reported is less than the total run-time duration of the input videos. Or to put it another way, a system with relative time as reported on the ActEV19 Sequestered Data Leaderboard (SDL) of at most one.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Mar 01, 2020
Accession Number
AD1155045

Entities

People

  • Carlos D. Castillo
  • Joshua Gleason
  • Rama Chellappa

Organizations

  • University of Maryland

Tags

Communities of Interest

  • Air Platforms
  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Accuracy
  • Algorithms
  • Aspect Ratio
  • Classification
  • Clustering
  • Computer Vision
  • Computers
  • Detection
  • False Alarms
  • Images
  • Networks
  • Pattern Recognition
  • Recognition
  • Sampling
  • Test And Evaluation
  • Training
  • Validation
  • Vehicles

Fields of Study

  • Physics

Readers

  • Computer Science/Computer Engineering/Data Science/Digital Signal Processing.
  • Exercise and Sports Science.
  • Sensor Fusion and Tracking Systems.