Research On Vehicle Object Recognition Method Based On Image Multi - Feature

As the new traffic law execution in2013, punishment for illegal vehicle license plate increased At present, enforcement of traffic police is mainly by vehicle license plate which is recognized by the vehicle license plate automatic recognition system. In order to avoid punishment, drivers often shielding, fouling, forging vehicle license plate, and the vehicle license plate automatic recognition system failed to do the recognition. Therefore, it is a difficult problem to identify the vehicle object which can’t be recognized by the vehicle license plate automatic recognition system. So it has a broad application prospect, and it’s also a hot research field crossing with computer vision, image processing and pattern recognition and other subjects. In this context, this thesis explored the classical statistical pattern recognition and image processing theory, and studied the vehicle object recognition feature extraction and recognition method in detail. Solved the problem of vehicle body color and type recognition, and have a significance in the fighting illegal vehical filed. This thesis proposed a solution that aimed at illumination, noise, scale-similarity situations. The main contents and innovations are as bellow:1. Establishment of standard models of feature model library. The standard models of feature model library is a basic work of this research, this thesis established the library from three aspects: physical features, structure features and vehicle photos, Including85%of micro cars detail information.2. Vehicle detection and color recognition. Through the average background modeling, proposed dynamic histogram threshold shadow elimination for object locating, can prevent the influence brings by shadow and other external environment. Improved the traditional vehicle color recognition method based on color space, and studied the brightness adjustment in unevenly distributed illumination situation. The improved algorithm can eliminate the interference from the light, and improve the accuracy of vehicle color recognition in the light changes.3. Research of vehicle classification and recognition based on multi feature fusion. Firstly, using the improved Adaboost method to locating the car face. The SIFT, SURF invariant features were extracted, and proposed an improved matching strategy, combined with the invariants of the car face region segmentation logo and headlights, and the textures of the radiator fences, using the standard models feature data, an adaptive step model classification algorithm of multi feature fusion are proposed. Statistical results showed that the fusion of multiple features can improve the recognition rate of the vehicle.