I. Introduction
The development of intelligent transportation systems (ITS) brings new technologies to solve traffic issues, including congestion, accidents, delays, and pollution. In the applications of ITS, such as traffic light control and intelligent vehicles, there is an increasing demand for traffic data extraction. To extract traffic data automatically and timely, vision-based vehicle recognition is an essential and challenging task. It collects vehicle physical attributes and vehicle traveling data for traffic management and control in parallel transportation systems [1] and has high industrial potential in advanced driver assistance systems [2] and autonomous vehicles [3].