Scene complexity and invariant feature based real-time aerial video registration algorithm

Tao Yang, Yan Ning Zhang, Xiu Wei Zhang, Xin Gong Zhang

Research output: Contribution to journalArticlepeer-review

13 Scopus citations

Abstract

Real-time and robust image registration is the premise and key technology of aerial video stabilization, panorama stitching and ground moving target detection and tracking. This paper presents a novel scene complexity and invariant feature based aerial video registration algorithm. The main characteristics of the proposed method include: (1) Based on analyzing the key difficulties and challenges of aerial video registration, several new methods are presented to realize fast and effective video registration under various real scenes, include integral image based fast image scale space generation, scene complexity based feature number controlling, and statistical error distribution of correspond features based cascade filtering. (2) Through combining the multi-scale Harris corner detection, SIFT (Scale Invarianu Feature Transform) feature description, and the RANSAC (Random Sample Consensus) based frame geometry transformation parameters estimation, the proposed algorithm achieves satisfied rotation, scaling, brightness invariance and accuracy of registration. Experiment results show that the proposed algorithm carries out real-time and precise image registration under complex conditions with change of scene, large image translation, scaling and arbitrary rotation, and the average processing speed for a resolution of 320x240 unmanned aerial video sequences achieves 20. 7fps.

Original languageEnglish
Pages (from-to)1069-1077
Number of pages9
JournalTien Tzu Hsueh Pao/Acta Electronica Sinica
Volume38
Issue number5
StatePublished - May 2010

Keywords

  • Aerial video registration
  • Invariant feature detection and matching
  • Video surveillance

Fingerprint

Dive into the research topics of 'Scene complexity and invariant feature based real-time aerial video registration algorithm'. Together they form a unique fingerprint.

Cite this