Background Normalization and Texture Pattern-Based Video Segmentation for Visual Tracking

Abstract

Visual tracking is the task of estimating the path of a target object in each frame of the video. Various tracking approaches are developed for tracking the position of the target. However, the tracker may fail to find the correct position, if the appearance of the target is similar to the background. This paper proposes a background normalization technique based on the textural pattern analysis to verify the matching of the features for the target region analysis. In this work, a novel model of background clustering is presented by using Multi-Weighted Chain Prediction (MWCP) algorithm for the uneven background. A Neighborhood Differential Binary Pattern (NDBP) is proposed to extract the texture for suppressing the shadow pixels in the image frame. From this equalized frame of the given video, the frame is split into several grids. From the grid format frame, the histogram features of the targeted frame are extracted, and each grid in that frame is classified. The Multi-Grid Weighted Classifier (MGWC) algorithm is used to find the matching of the grids to separate the background and foreground. This type of visual tracking system is robust over the sudden illumination changes and dynamic background. The experimental result proves that the proposed approach yields better precision, recall and F-score performance than the existing tracking approaches.