A study of household object recognition using SIFT-based bag-of-words dictionary and SVMs

Abstract

In the era of computational intelligence, computer vision-based techniques for robotic cognition have gained prominence. One of the important problems in computer vision is the recognition of objects in real-time environments. In this paper, we construct a SIFT-based SVM classifier and analyze its performance for real-time object recognition. Ten household objects from the CALTECH-101 dataset are chosen, and the optimal train-test ratio is identified by keeping other SVM parameters constant. The system achieves an overall accuracy of 85% by maintaining the ratio as 3:2. The difficulties faced in adapting such a classifier for real-time recognition are discussed.