Context based object detection from video

  • Authors:
  • Lucas Paletta;Christian Greindl

  • Affiliations:
  • Joanneum Research, Institute of Digital Image Processing, Graz, Austria;Joanneum Research, Institute of Digital Image Processing, Graz, Austria

  • Venue:
  • ICVS'03 Proceedings of the 3rd international conference on Computer vision systems
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

The past few years have seen a dramatic request for semantic video analysis. Object based interpretation in real-time imposes increased challenges on resource management to maintain sufficient quality of service, and requires careful design of the system architecture. This paper focuses on the role of context for system performance in a multi-stage object detection process. We extract context from simple features to determine regions of interest, provide an innovative method to identify the object's topology from local object features, and we outline the concept for a correspondingly structured system architecture. Performance implications are analysed with reference to the application of logo detection in sport broadcasts and provide evidence for the crucial improvements achieved from context information.