Research On Pedestrian Detection Technology Based On Support Vector Machine

Pedestrian detection is the use of computer vision technology to determine whether there are pedestrians in images or videos,and locate them accurately,and detect the status and posture of the pedestrians when necessary.The technology has been widely used in vehicle assisted driving system,intelligent video surveillance,intelligent transportation and other fields,and has become one of the research hotpots in Machine Learning(ML).The methods of modern pedestrian detection are mainly divided into two categories.The first is based on background modeling,The process of this method is to split the image first,then extract the target,then extract the feature,and finally classify it.This method is sensitive to environmental changes,such as lighting and weather.The other is based on statistical learning to construct a classifier which is a more commonly used method.This method is based on a large number of learning samples.In this paper,the classifier is designed based on second method,and pedestrians are used as samples to detect the recognition effect of the classifier.The traditional algorithm of pedestrian detection is based on Histogram of Oriented Gradient(HOG)and Support Vector Machine(SVM).HOG algorithm is used to extract the feature of the pedestrian,and SVM is used as the classifier.This method is the traditional method for pedestrian detection.The shortcoming of this method is that the accuracy of recognition is low and the time consuming is long.It is difficult to meet the requirements of the accuracy and real time of the modern pedestrian detection.In this paper,the classifier is improved on the basis of traditional algorithm.HOG algorithm is used as the feature extraction algorithm,and the Principal Component Analysis(PCA)algorithm is used to reduce the dimension of the feature vectors,which reduces the time consumption and improves the real-time performance.This paper proposes a combination kernel function as the kernel function of SVMclassifier.The combination kernel function is a linear combination of polynomial kernel function and Gauss kernel function,which makes the classifier has global characteristic,also has a local characteristic.Because of the kernel function,the classifier have influence on the nearby data,and also can act to far away from the test point data at the same time.This paper set relaxation variables,and give up the rigid classification of the outliers in the near linear separable conditions,so that the optimal classification hyperplane can be avoided to move because of the separation of group points,thus obtaining larger geometric intervals.The penalty factor C is cited to adjust the balance between the maximum geometric interval and minimum slack variable.C represents the degree of attention to outliers.The selection of C directly affects the generalization ability of classifier.only the C which is fit for the kernel function can make the classifier get the better generalization ability.In the optimization process above,the selection of the C,the combination coefficient of the combined kernel function and the parameters of the kernel function all can affect the accuracy of the classifier.For better classification effect,this paper combined Genetic Algorithm(GA)and k-fold cross cross validation validation(K CV)to optimize the parameters.Finally,the optimized classifier is used as a weak classifier of AdaBoost.The final strong classifier is got by stratifying iteration and changing the weights,the high weights focus on the the sample of difficult classification,which are easier to be misclassified.Finally,in order to verify the performance of the classifier,this paper contrast the recognition rates of the classifier of the paper,the classifier based optimized SVM,the classifier based on traditional method.By the result of comparison,the increase of the classifier in this paper in pedestrian detection can be seen obviously.