Recognition Of Car Based On Loacal Feature

The recognition of car type has a variety of potential applications in the real world such as automated surveillance, public security, intelligent vehicles, and traffic monitoring and control system. Due to the complex background and the various of car appearance, car recognition is still unresolved problem. So it is very necessary to do some research work on car recognition.The recognition system proposed in this paper is based on two steps: First, Detecting the car and getting the region of car face is the first step to performing car recognition, the region of car face image make sure that the following local feature matching is done with only car features and not background features. Second, SURF or SIFT features are previous extracted from a set of reference images (only one per car class) and stored in databases. A new input car face image is matched by individually comparing each SURF or SIFT feature from the new image to the database. For recognition, the image in the database which has the largest number of matched features is used to measure the type of the car. In our experiment, the SIFT local feature based algorithm yielded a recognition rate of 75.47% and the SURF local feature based algorithm yielded a recognition rate of 78.03%when tested on about 800 images containing 48 different kinds of car type.Gabor was used as a tool for texture analysis of the vehicle region, shape-context feature was used to represent the vehicle shape information. Gabor feature combined with the SURF yielded a recognition rate of 81.27%, the shape context feature combined with SURF yielded a recognition rate of 82.54%.Experimental results show that a higher recognition rate can be obtained by using different features, all of which can contribute useful matches and improve overall robustness.