The Construction of Activity Modeling for the Video Content Analysis Based on Semantic Concept Using Trajectory Representation

摘要

This paper presents the construction of activity modeling for video content analysis based on semantic concept using trajectory representation. In video content, a semantic concept will be interpreted according to the related position between the forward object and the background object, with conformity the semantic concept invariance with the forward object in its activity scope. A special property of SVM is known as a maximum margin classifier, which simultaneously minimizes the empirical classification error and maximize the geometric margin. The activity scope of each object and support vector will be decided by SVM decision boundary and principal axis analysis which is used to construct a hyper-plane projection line with minimum inertia. In this paper, we propose a novel method to find the maximum and fair activity scope for every forward object in a frame, which will be used to analysis the semantic concept of video content. The semantic concept of video content will be transformed into a two-dimension trajectory representation where a set of trajectory breakpoints will be acquired using the SVM, which will dismember the semantic concept of video content in the trajectory representation. We are also proposing three types of activity modeling in video content. Experimental results on a set of video sequences have shown the effectiveness of analyzing the video content using the proposed activity modeling and they can be easily obtained from the measurement of the proposed method. Experimental results will show that the performance of this proposed method is excellent when compared with that of other methods.