Gabor filtering for feature extraction in real time vehicle classification system

Vehicle classification problem is one of challenges in Intelligent Transportation System (ITS). Numerous approaches have been submitted to handle this problem. Real time environment condition and limitation make it more intriguing. In this paper we focus on real time vehicle detection, feature extraction, and classification for multiple object using a single stationary camera. Even though numerous approaches have been proposed, there is still no study that is robust in every condition (object collision, lighting, vehicle appearance, vehicle type, video resolution, etc). These conditions stimulate this study more challenging. We developed a real time system to conduct classification process using three main steps: multiple vehicle detection using Gaussian Mixture Model with Hole Filling Algorithm (GMMHF); Gabor kernel for Feature Extraction; and multi-class vehicle classification. In this paper five classifiers were applied to compare the classification process. Proposed scheme shows that our system successfully detects and classifies the objects in the real time condition. The highest accuracy is 93.36% that is obtained using 18 features built by ten Gabor kernel combinations with Random Forest classifier.