Login | Register

Video object detection using fast and accurate change detection and thresholding

Title:

Video object detection using fast and accurate change detection and thresholding

Su, Chang (2007) Video object detection using fast and accurate change detection and thresholding. Masters thesis, Concordia University.

[img]
Preview
Text (application/pdf)
MR28929.pdf - Accepted Version
4MB

Abstract

Video object detection is an important video processing technique. Change detection and thresholding based video object detection techniques are widely used due to their efficiency. However, change detection and thresholding in real-world video sequences is challenging due to the complexity of video contents and of environmental artifacts. This thesis proposes a color-based change detection and a video-content adaptive thresholding method for accurate and fast video object detection. The proposed color-based change detection algorithm is based on the YUV color model, which has been proved as the most effective color model for object detection. First, frame-differencing is carried out in each channel of a video frame. Then, the pixel intensities in both gray-level channel Y and the color channels U and V of the difference frames are statistically modeled. Second, based on the statistical model of the gray-levels in Y channel, an entropy-based blocks-of-interest scatter estimation algorithm is proposed for locating the frame blocks potentially containing moving objects; and based on the statistical models of the color intensities in color channels, a statistical model of the maximum-intensity between U and V channels are obtained. Third, significance test is applied to the detected blocks-of-interest in both gray-level channel and color channels based on the gray-level statistical model of Y channel and the maximum-intensity statistical model of U and V channels. The gray-levels of the non-significant pixels in Y channel but significant in the U or the V channels are then compensated according to their significance probabilities in the color channels. Finally, change masks can be obtained by a thresholding algorithm. The proposed thresholding algorithm for change detection is based on a change region scatter estimation algorithm and a video-content assessment algorithm to detect the empty frames and estimate the strength of local unimportant changes. According to the proposed video-content assessment, the global threshold of a difference frame is discriminatively computed. For an empty frame, a noise-statistic based thresholding algorithm with a low false alarm is applied to obtain the threshold. Otherwise, the global threshold is obtained by an optimum-thresholding based artifact-robust thresholding algorithm. Experimental results show that (1) with the support from the scatter estimation of the blocks-of-interest, the proposed change detection algorithm is efficient and robust to multiple video contents; (2) the proposed thresholding algorithm clearly outperforms the widely used intensity-distribution based thresholding methods and more efficient and more stable than the state-of-the-art spatial-property based thresholding methods for change detection; and (3) the video object detection technique consisting of the proposed change detection and the proposed thresholding algorithms is robust to artifacts and multiple video contents, and is especially suitable for real-world on-line video applications such as video surveillance

Divisions:Concordia University > Gina Cody School of Engineering and Computer Science > Electrical and Computer Engineering
Item Type:Thesis (Masters)
Authors:Su, Chang
Pagination:xvii, 106 leaves : ill. ; 29 cm.
Institution:Concordia University
Degree Name:M.A. Sc.
Program:Electrical and Computer Engineering
Date:2007
Thesis Supervisor(s):Amer, Aishy
ID Code:975326
Deposited By: Concordia University Library
Deposited On:22 Jan 2013 16:06
Last Modified:18 Jan 2018 17:40
Related URLs:
All items in Spectrum are protected by copyright, with all rights reserved. The use of items is governed by Spectrum's terms of access.

Repository Staff Only: item control page

Downloads per month over past year

Back to top Back to top