Vehicle classification in congested traffic based on 3D point cloud using SVM and KNN

Abstract

The vehicle classification in congested traffic is a big challenge due to the difficulty to segment packs of different vehicles that stand still next to each other or travel at a very low speed. In this work, a low-cost vision system was designed and built to acquire the image and to generate 3D point cloud to be used as input for the classification process. The vehicle classification uses machine learning K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) with radial basis function kernel to classify two types of vehicle which are car and motorcycle based on 3D point cloud. The processing of the training data and test data can be divided into filtering, segmentation, tracking, and feature extraction, respectively. The extracted feature vectors are then used for both KNN and SVM classifiers. The results show that the proposed performs well even in high congested traffic with a mix of both vehicle's type. This can be seen from the TPR for car classification from both KNN and SVM which is relatively high (KNN=95.8% and SVM=95.8%) compared to other existing systems. In case of motorcycle classification, the SVM classifier performs better compared to KNN in all three different traffic conditions.