Spatial-temporal pyramid matching for crowd scene analysis

Hanhe Lin, Jeremiah D. Deng, Brendon J. Woodford

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

Abstract

Crowd scene analysis has caught significant attention both in academia and industry as it has a great number of potential applications. In this paper, we propose a novel spatial-temporal pyramid matching scheme for crowd scene analysis. Video segments are represented as concatenated histograms of all cells at all pyramid levels with corresponding weights, which reect corresponding matches at finer resolutions are weighted more highly than that found at coarser resolution. Using the classical stochastic gradient descent method, we also propose an online one-class support vector machine algorithm for online anomaly detection scenarios. Extensive experiments have been carried out on two benchmark datasets and the comparison to the state-of-the-art methods validate the advantages of our approach.

Original languageEnglish
Title of host publicationProceedings of MLSDA 2014
Subtitle of host publication2nd Workshop on Machine Learning for Sensory Data Analysis
EditorsAshfaqur Rahman, Jeremiah D. Deng, Jiuyong Li
Place of PublicationNew York, NY
PublisherAssociation for Computing Machinery (ACM)
Pages12-18
Number of pages7
ISBN (Electronic)978-1-4503-3159-3
DOIs
Publication statusPublished - 2 Dec 2014
Event2nd Workshop on Machine Learning for Sensory Data Analysis - Gold Coast, Australia
Duration: 2 Dec 20142 Dec 2014

Conference

Conference2nd Workshop on Machine Learning for Sensory Data Analysis
Abbreviated titleMLSDA 2014
Country/TerritoryAustralia
CityGold Coast
Period2/12/142/12/14

Keywords

  • Anomaly detection
  • Crowd event recognition
  • Online one-class support vector machine

Fingerprint

Dive into the research topics of 'Spatial-temporal pyramid matching for crowd scene analysis'. Together they form a unique fingerprint.

Cite this