Vehicle Logo Detection Using Sliding Windows with Sobel Edge Features and Recognition Using SIFT Features

Abstract

Automated traffic monitoring is one of key component for a smart city due to its high efficiency and availability compared with human based monitoring. License plate recognition systems are widely used; however, information such as vehicle make are also required and still lack a practical method. Therefore, this paper proposes a practical method for vehicle logo detection and recognition in concern with a real-life surveillance situation. Instead of locating a logo from an entire image as several published proposals, sliding window method is proposed to locate candidate areas where a vehicle logo resided. The area is identified by the maximum number of Sobel edges compared among the candidate areas. The logo in the identified area is recognized using the SIFT based features and a Nearest Neighbor classifier. The proposed method is experimented with real-life traffic video surveillance images. The images are low resolution under various daylight condition. The proposed method is trained and experimented with 3,176 images of nine vehicle makes. The proposed method is assessed using confusion matrixes and shows overall accuracies in range of 85%.