Motorbike Detection in Urban Environment

Chi Kien Huynh, Tran Khanh Dang, Thanh-Sach Le

Abstract: Vehicle detection is one of the key components in vision-based intelligent transportation system (ITS). Having an accurate detector in difficult and varied environments is a prerequisite for a usable ITS. Many existing methods were tested on environments where there are a lot of four-wheelers and the general driving is in order. However, the situation is different in Vietnam and similar countries where motorbike is the major vehicle choice. This makes traffic much more unpredictable and difficult for vision detectors. In this work, we applied different detection methods to analyze their performance on our traffic datasets. The results show that deep neural network approach offers better accuracy than the others. They also prove that in circumstances that are under control, these methods can be readily integrated in practical traffic systems.