Visual category recognition for the improved storage and retrieval performance of the CCTV camera system

摘要

In this paper, we propose a category level object recognition system for the efficient use of CCTV cameras in terms of storage and retrieval. We investigate the performance of the proposed approach by using four different classifiers. More specifically, we considered image sequences with cars, bikes and pedestrian as our three targeted object categories for classification and ultimately efficient storage and retrieval with reference to our CCTV cameras system. We utilized Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), K-Nearest Neighbors (KNN) and Cartesian Genetic Programming (CGP) algorithms for the considered object categories classification. The Linear Discriminant Analysis (LDA), KNN and Support Vector Machine (SVM) are Statistical algorithms while Cartesian Genetic Programming (CGP) is Evolutionary Algorithm. More specifically, we utilized the standard “Caltech 101” dataset for investigating the performance of our proposed classifiers. Scale Invariant Feature Transform (SIFT) has been used to extract the scale, orientation and translational invariant features from the considered images which are input to the classifiers. Our empirical results show that in most of the cases, the results of LDA and SVM are relatively the same. To be specific, LDA gives an average accuracy of 85.3% and SVM 83.6%. Similarly, KNN gives an average accuracy of 74.6% while CGP outperforming the three gives accuracy rate of 89%.