Experimental Study: Influence of Feature Extraction in Objects Multiclassification

Moving object multiclassification is an important step in a smart video surveillance system. However, the relevance of its results depends not only on the classifier but also on the used data. Instead of the edge characteristic of the moving targets, representative characteristic features are extracted and used in the classification. In this paper, firstly, we present a comparative and experimental study of the effectiveness of several features and their influence on the classification accuracy. Secondly, we evaluate the performance of SVM-Multiclass, Artificial Neural Network (ANN), and K-nearest neighbor (KNN) with the extracted features. Also, we propose the use of feature fusion and the experimental results show that it improves the performance of the used classifiers and increase the classification accuracy up to 94%.