The Research Of License Plate Location Algorithm Based On Combining Multi-feature

License Plate Recognition System(LPRS)as an intelligent and automated traffic management tools, which greatly facilitated our lives. License plate location is the key part of LPRS, which directly affects the performance of the entire system. Most of previous license plate location algorithms are not robust enough in complex environments, have high false alarm rate. So the further research of license plate location algorithm is of great significance.License plate location based on machine learning methods have more advantages to handle complex environments and multi-plate, becoming a research hotspot in recent years. License plate location algorithms based on Cascade Ada Boost were used more often, but have high false alarm rate, and need a lot of samples. In this paper we improve the feature extraction method of traditional Ada Boost method, greatly enhance the classification ability of Ada Boost classifier. Train single-stage strong Ada Boost classifier to eliminate background areas quickly, construct three features and train combining SVM classifier for plate precise location. Finally design a license plate location system based on combining Multi-feature. The main work is as follows:Firstly, introduce a new stationary dyadic wavelet transform, and proposed the improvement of extracting haar feature using wavelet coefficients. Train Ada Boost classifier in our training samples, experiments show that our method has stronger classification ability and robustness compared to traditional haar feature extraction based on image pixel directly.Secondly, combine SVM classifiers based on Multi-feature, construct three features based on wavelet coefficients, train three SVM classifiers and combine them. Experiments show that the three features have strong classification ability, through the combining, the final classifier greatly reduces the false alarm rate while maintaining a high location rate.Thirdly, design classifier cascade framework, use the trained classifiers, develop a multi-feature combined license plate location system.This paper uses the combining Ada Boost learning method, support vector machine(SVM) method, etc. Analyze the characteristics and advantages of the various machine learning methods, combined them effectively. Train a combined classifier using multi-feature, use the combined classifier to experiment on the test set for license plate location. Our method achieved a higher location rate and false alarm rate, and robust to different environments.