Motion Data-Driven Model for Semantic Events Classification using an Optimized Support Vector Machine

摘要

The spatio-temporal constraints that accompany video data types are one of the unique characteristics of video information. The importance of the temporal constraints has led to recent efforts to incorporate them in video events representation, indexing and retrieval. To support the classification of a given video event, we propose a data-driven model which utilizes the motion information to enhance event classification performance. Kernel-based methods have become popular in multimedia classification tasks. However, in order to use them effectively, several factors that hinder accurate classification results, such as feature subset selection and kernel parameters, must be addressed through the use of heuristic-based techniques. Here, we present a novel approach to enhance the performance of support vector machine based on a search method. The latter relies on the simultaneous optimization of: (i) the feature subset, (ii) the instance subset and, (iii) the SVM kernel function parameters, with genetic algorithms. Experimental results on a collection of sports videos show that this method significantly improves the classification accuracy of conventional SVM based techniques.