Comparison of Background Subtraction Methods on Near Infra-Red Spectrum Video Sequences

Background subtraction methods are used to detect foregrounds objects in video sequences. However, a lot of parameters of video sequence could complicate this process. Like noise, moving trees, rain, wind etc. Most popular methods are based on Gaussian mixture models (GMM). Four methods based on GMM were used: GMG, KNN, MOG, MOG2. Comparison is realized by using twenty video sequences captured in near infrared spectrum. Each video sequence has one or more moving wild mammals. On twenty randomly selected frames the moving objects are manually segmented for each video. Manual segmentation is done by group of people. Then, results from background subtraction methods are compared opposite to human segmentation by using brute force matcher and were improved by using Radon transformation. From results is obvious the KNN has the biggest similarity opposite to human segmentation. The method with the best correlation opposite to human segmentation will be used in near future for animal detection purpose.

Language

  • English

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 01644823
  • Record Type: Publication
  • Files: TRIS
  • Created Date: Jun 26 2017 3:18PM