Vehicle Classification using Support Vector Machines and k-means Clustering

Abstract

This investigation involves the combination of support vector machines (SVM) and k-means clustering implemented on radar signal. SVM classifier based on k-means algorithm has an advance for classifying unlabeled data. To classify the vehicle types automatically, the combined method is implemented with radar signals. This paper proposed a classifier training algorithm based on SVM and k-means clustering as follows (i) using the k-means algorithm to label the input feature data extracted from radar signal into two subsets, (ii) train SVM with labeled data, and (iii) classify unidentified radar signal into large vehicle or small vehicle with the trained classifier. Training features of radar signal includes (i) signal volume and (ii) sum of signal variations, both in frequency domain. These features are taken as system input, while vehicle types as system output. The proposed algorithm is implemented and demonstrated with real FMCW radar signals. With the numerical experiment, satisfying result is obtained.