Real-time taxi detection for embedded system

This paper presents a real-time taxi detection algorithm for embedded system. The detection is achieved by setting detection areas, license plate detection, vehicle roof detection, vehicle roof feature extraction and vehicle roof classification. In other words, a frame is extracted from the video and transformed into a rectangle image. After that, the system locates the license plate and followed by the vehicle roof. Then the features of the vehicle roof and the vehicle color are extracted. Finally, the features are input into a regression function, which is obtained from previous training data, to determine whether the vehicle is a taxi. These features are easy to extract so that the whole process is simple and robust. Due to the small cost of time, this method is suitable for embedded system.