ANALYSIS OF VIDEO SEQUENCE USING INTELLIGENT TECHNIQUES

Abstract

With the increase of advent of technology and demand of society, video sequence analysis based systems are becoming reality of various applications and active research area of the computer vision. The wide range of applications includes video based intelligent surveillance, motion analysis, indexing of video, web based video filters, human computer interaction, human activity recognition, object tracking, smart interactive televisions, animations and special effects in movies, sports analysis and so forth. The main building blocks of a video sequence analysis system consist of pre-processing, features extraction and representation, and classification. newlineIn view of the various applications of video sequence analysis, this thesis investigates human activity recognition approach based on human silhouettes. Human silhouette is the basic information unit for representation and recognition of human activity is. To achieve higher recognition accuracy of human activity, a three step methodology is devised: newlineand#61607; First step is extraction of human silhouette using texture based foreground segmentation; newlineand#61607; The second step is extraction and representation of features, which is done by two major approaches: one is the grids and cells based and the other is computation of spatial distribution and rotation of human silhouette; newlineand#61607; The third step is classification of human activity performed by various state-of-art classifiers. newlineiii newlineIn the activity recognition methodology based on human silhouette and cells, the key poses of the human silhouettes are selected based upon the high energy principle. These key poses are divided into various cells and further features are computed. The computed features are then represented in such a manner that the spatio temporal information of the human silhouette is maintained. The represented features are used to form a feature vector and these feature vectors are classified using linear discriminant analysis (LDA), K-nearest neighbour (K-NN), and support vector machine (SVM).