Title

Improving the Teknomo-Fernandez background image modeling algorithm for foreground segmentation

Date of Award

2015

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Computer Science

Department

Information Systems & Computer Science

Abstract

  • The accuracy of object detection depends on the background modelimage of the scene used in the background subtraction process. The TeknomoFernandez(TF3) algorithm is an efficient background modeling algorithm thatgenerates a good background model image quickly. It approximates the modelbackground image by using a tournament-like strategy and incorporates multilevelprocessing. This study aims to improve the accuracy of the TF3 algorithm for betterbackground modeling and, ultimately, more accurate foreground segmentation. Animproved TF3 variant implements intrinsic and extrinsic improvement techniques thatinvolves a selective frame range sampling for the background maintenance step and aseries of morphological operations. The performance was evaluated using theWallflower dataset and compared against existing state-of-the-art backgroundsubtraction algorithms. Empirical results showed that the improved TF3 has a significantimprovement of 90.72% over the original TF3 in terms of accuracy. Moreover, it wasshown to be adaptive to slight changes in the scene with its best performance on theLS, TD, WT and FA test videos. Compared to current state-of-the-art, the improved TF3generally works well across all test scenes and is the best in terms of accuracy with theleast number of error count for all test videos. Empirical results show that the TF3algorithm is one promising technique for real-time object detection.

Comments

The C7.A34 2015

Share

COinS