A Vehicle-logo Recognition System Based On Smart Phone

In recent years, quantity of cars in China grows rapidly, transportation is facing more and more pressures and challenges, intelligent transportation is the development direction of modern transportation industry. Vehicle recognition based on license plate recognition is a critical technique in intelligent transportation system, which has significant commercial value and realistic significance. Combining vehicle-logo recognition technique and license plate recognition technique will obtain more accurate vehicle information, and improving the recognition rate and robustness of vehicle recognition system.In this paper, vehicle logo location method is divided into vehiclelogo location stage and vehicle-logo classification stage. In vehicle logo location stage, a coarse to fine vehicle-logo location methods based on location of license plate is used. Firstly, license plate is located based on its color feature in RGB color space and HSV color space. Secondly, a coarse vehicle-logo location was obtained based on the position relationship between vehicle-logo and license plate. Finally, detecting edge of coarse region by conjuncting horizontal and vertical edges, which will eliminate the edge of the heat grid, the noise is filtered out by open operator of mathematical morphology. The edge projection and gravity center method is used to location accurate vehicle-logo location, starting from gravity center of edge image, locating boundaries in vehicle and horizontal edge projection. In vehicle-logo recognition stage, firstly normalizing the vehicle-logo image to 32 × 32, then a 144-dimensional HOG feature is extracted, vehicle category can be obtained by SVM linear classifier according to HOG feature.Experiments show that the vehicle-logo recognition method proposed in this paper has faster rate in processing, suitable for real-time systems.Using above vehicle-logo recognition method, a vehicle-logo recognition mobile application on Android platform is developed for the first time, with OpenCV4 Android SDK. In the application, OpenCV Manager is used to manage the OpenCV library, image is captured from camera through CvCameraViewListener interface. The results shows that vehicle-logo recognition method in this paper is efficient and effective, the average processing time less than 0.2 seconds, mobile application, which meets the real-time requirements.