Research On Vehicle Detection Method Based On HOG Feature Extraction

In recent years,with the developement of our country's economic,the automobile manufacturing is developing rapidly.The traffic system can't run normally because of too many cars.The traffic intelligent transportation system has become a hot spots to improve the current circumstances.In this thesis,the vehicle detection method based on feature extraction is studied deeply.Including the feature extraction and classifier classification.Features includes HOG features,SIFT features,Haar features,Harris features,etc;Classifier includes SVM classifier,Adaboost classifier.The traditional vehicle detection method has a high false detection rate and a low rate in the complex environment.Which can not meet the actual detection requirements and needs to be improved.In order to overcome the shortcomings of HOG feature description,the key bin and the amplification factor are introduced into the traditional HOG feature extraction method.At first,the key bin of the vehicle image is extracted by the average gradient difference of the sample set image.Then the key binl of the HOG feature is amplified by the amplification factor.So the HOG features of local amplification are obtained.In order to enhance the expression of vehicle texture information,the LBP feature is mixed.And the mixed feature is more powerful.The detection rate can be improved and the false recognition rate of the sample set can be reduced.In order to solve the problem of high feature dimension and long computation time,the additive crossed kernel SVM is selected for classification and detection.The classifier can effectively reduce the time of classification and detection,and finally use sliding window to detect.The method was tested by using UIUC vehicle dataset.The experiment results show that the proposed method is good at detection.It has a lower detection rate and a higher detection rate than the traditional HOG + SVM method.