The Research Of Pedestrian Detection Based On Multiple Classifier

As the number of cars increasing rapidly in China,road safety has attracted more and more attention.Pedestrians are the main participants in the traffic,but their safety often can't be guaranteed.As a part of auxiliary driving system,the pedestrian detection system controls the car from a way of active pedestrian detection,playing a role of protecting the pedestrians.But at present,the pedestrian detection system often can't satisfy the requirement in aspect of speed and accuracy.Therefore,it is necessary to design a more rapid and accurate pedestrian detection system in order to meet the need of fast and efficient pedestrian detection.In this thesis,intensive research has been made regarding speed and accuracy to find the methods and ways to improve the efficiency of detection.Moreover,it proposes rapid pedestrian detection algorithm and accurate pedestrian detection algorithm.Finally,the research integrates the results of rapid pedestrian detection research with the results of accurate pedestrian detection research.Combining with the advantages of both,we propose a cascade of multiple classifier way to realize the fast detection of pedestrians.Our main research contents include:1)Design a quick pedestrian detection algorithm based on LBP.Algorithm use LBP operator to extract the pedestrian's features simply and efficiently,using integral image method to accelerate the calculation of pedestrian detection,using support vector machine(SVM)modeling the features of pedestrian,and ultimately achieve the purpose of rapid detection of pedestrian.2)Design an accurate pedestrian detection algorithm based on Regionlet.Algorithm uses stochastic prediction method to predict the area that pedestrian may arise,and divides the area into small areas,names regionlet.Extract the feature from the area of regionlet.After processed all the features extracted from regionlet,we obtain the feature of big area.Because of less deformation of pedestrian in smaller areas,it has a better adaptability to deform and higher robustness to detect the pedestrian.Using adaboost classifier to detect the pedestrian areas can reduce the false detection rate effectively.In this way,it improves the accuracy of pedestrian detection.3)Build pedestrian detection system based on Multi-classifier.The system combined the fast classifier and the accurate classifier.First,we use fast classifier to do a rough extraction on the detected image and then the filtrating results are sent to the second classifier to do a second detection.Because of reducing the amount of information to be detected,so that it could improve the detection speed and achieve the purposes of pedestrian detection more quickly and accurately.The experimental results show that the research results can effectively improve the pedestrian detection speed and accuracy,providing a useful reference and help for the realization of real-time pedestrian detection.