Analysis of Various Clustering Algorithms to Enhance Bag-of-Visual-Words for Drowsiness Prediction

Bag-of-Visual-Words is a technique used to create image vocabularies which describes the best image features. The construction of visual vocabulary is done using various clustering techniques. This work concentrates on various clustering techniques that are implemented on Bag-of-Visual-Words technique so that to analyse the accuracy of vocabulary creation. The clustering techniques such as K-means, Mini-batch K-means, Mean-shift, DBSCAN and OP-TICS are implemented individually to record the efficiency of the model. Features from the input images are extracted using Scale Invariant Feature Transform(SIFT) matched with Fast Library for Approximate Nearest Neighbors(FLANN). Drowsy images are classified based on the occurrence of the visual words. The comparison result indicates that the OPTICS clustering algorithm works well with Bag-of-Visual-Words to output an accuracy rate of 79.01%.