A video target tracking using shadow suppression and feature extraction

In real-time application, most of the research is focused in video tracking system. Since this is mainly used for the application of robotics, surveillance tracking, human to machine interface, etc. to extract the target status. There are more number of video tracking system to identify the targeted region from the frame of given video. Since the performance of tracking can affect by sudden illumination changes, shadowing effect and uneven background. In moving object detection system, it suffers from dynamic background changing and shadow effect present in the video frames. Due to this, tracking of targeted region may misclassified and results in false detection of moving objects. To rectify this problem there are many techniques to detect and eliminate the shadow from frames like K-Means clustering, Fuzzy C-Means, etc. which are to segment both foreground and background from each frames. Then they remove/suppress shadow region and track the target. Since in that methods, segmentation is based on non-changing background of surveillance area. To rectify this problem, we propose a novel background normalization technique which is based on textural pattern analysis. In texture based system, there are several methods perform texture pattern extraction to verify the feature matching for target region analysis. In this work, we present a novel model of background clustering by using Neighborhood Chain Prediction (NCP) algorithm for uneven background. Here we also propose Differential Boundary Pattern (DBP) to extract texture of the frame for suppressing shadow pixels present in the frame. This is done by estimating lowest intensity present in that frame and predict the area by using DBP method and enhance the pixel to suppress shadow region. From this equalized frame of the given video, we split the frame into several grids. Then from that grid formatted frame, we extract histogram features of the targeted frame and provide classification for each grid in that frame. In that classification can be done by using Machine Learning Classification (MLC) method. This matched grid is consider as the tracked region and provide binary label to separate background and foreground. This type of visual tracking system is robust over sudden illumination changes and dynamic background by using the texture pattern analysis. Our Proposed work can be compare with existing segmentation approaches for the parameters of True Positive (TP), True Negative (TN), False Positive (FP), False Negative (FN), Sensitivity, Specificity, Accuracy, Correction rate, Positive likelihood and Negative likelihood for the tracking of targeted region.