A Vehicle Brand Classification Method Based On Computer Vision

Vehicle classification and recognition of road traffic system is the important part of intelligent transportation in modern city. Its application was used in parks and parking management, urban road traffic condition monitoring, and highway tolls, etc. computer vision, image processing, pattern recognition, machine learning techniques such as vehicle classification and recognition system, can play an important role in practical. In recent years, the advantage of easy to install and test range is a research hotspot and development trend.In this paper, in order to solve the deck with illegal vehicle identification problem, this paper proposes a vehicle brand recognition algorithm. the algorithm can extract the information of the front of the car’s face (mask), can achieve higher recognition accuracy, and deal with image in different angles and scale interference, and it has good adaptability for the change of illumination and the change of the weather conditions. This algorithm consists of two parts: interested area (ROT) positioning; Vehicle recognition based on PCA-NET. (1)The location of interested area (ROI):first to test the vehicle, this paper based on the symmetrical characteristic of vehicle detection method to determine the center of the plate, reusing the relative position of license plate and face in front of the car and the proportional relationship orientation interested area, And briefly analysis the selected region which was regard as a key principle region and its advantage of using it. (2) Basing on PCA-NET brand vehicle classification:PCA-NET the cascade principal component analysis (PCA), binary hash and block histogram. Study of principal components analysis using multilevel filter, and then use a simple binary hash indexes and pool operation and block histograms to implement the brand classification based on image. This article collected a large number of car image through the production and the Internet to download, and the test data on four groups of numerical experiments were tested under the condition of the day, under the condition of dark and light, with a shade, recognition rate under different perspectives. As you can see, from the result of the experiment for automobile front view recognition, recognition rate of this algorithm is reliable:daytime conditions is 94.17%.For multiple Angle of view under the condition of vehicle recognition, this paper reveals the Angle change on the impact of recognition.