The Developing of Dual Tires Detection Model of Two Axles Truck by Using 2D-PCA Feature Extraction and SVM as Classifiers

Abstract

Two axles truck is devided into two types i.e truck that uses single tire and dual tires at its back wheels. The use of dual tires at a truck will influnce its classification, so that it is needed a system to detect the use of dual tires. In this study, we develop a model to detect the occurance of dual tires at a two axels truck by using two steps 2D-PCA technique as the feature extraction and SVM as the classifier. In the feature extraction steps by using 2D-PCA, we use the values of precentage 95 %, 90%, and 85 %. While SVM use linear kernel, quadratic, cubic and RBF (sigma = 1, 5, 8, 10, 20, 30). By using the scenario, we obtain 81 models. We performed two phases of testing. The first testing phase measures the accuracy of the detection process without sliding windows. The second testing phase use sliding windows to detect the occurance of dual tires in an image. For the first phase testing, we use a database that consists of 552 dual tires images and 1284 non dual tire images with 150 x 150 pixels, and for the second phase testing, we used 30 images with 640 x 480 pixels. Based on the first phase testing, we obtained 10 best models to be used for second phase testing. The two stage 2DPCA method successfully reduced the data from 22500 dimensions of image vector to 36. The two phases testing conducted showed that the best kernels for detecting dual tires using SVM is the quadratic and the RBF kernel with the best accuracy of 93.3%.