Research On Pedestrian Detection Algorithm Based On Improved Feature Fusion

Feature based on pedestrian detection algorithm has always been a hot research project in the field of pedestrian detection.Among them,optimizing the feature extraction of image details and improving the accuracy by fusing a variety of different features have always been the research focus of researchers.This paper first optimizes the traditional feature extraction algorithm,then improves the feature fusion algorithm,and proposes a pedestrian detection algorithm based on improved feature fusion.The main research contents are as follows:(1)Firstly,the sample set is selected,and then the sample set image is preprocessed to prepare for the subsequent feature extraction.Through grayscale and size normalization,the influence of illumination and other factors on the sample image is minimized.This paper selects the best means to suppress noise by comparing the denoising effect of filtering methods.Then histogram equalization is used to improve the contrast of the picture.(2)We can understand the principle and some places that can be optimized by describing the commonly used feature algorithms hog algorithm and LBP algorithm for character detection and the classification algorithm of support vector machine SVM in detail,so as to pave the way for the optimization of subsequent feature algorithms and the selection of SVM parameters.(3)Firstly,through the experiment of hog feature and SVM classifier,the most suitable SVM kernel function and parameter setting are selected.Then this paper optimizes the algorithm of hog feature and LBP feature andimproves the extraction of edge gradient feature by hog feature,to reduce the time of extracting hog feature,and strengthen the description ability of LBP feature about texture part.(4)In view of the shortcomings of the single hog feature and the traditional multi-feature serial fusion method,an LBP-HOG feature fusion algorithm based on hierarchical fusion is designed,and the extracted features are dimensionally reduced by PCA in order to reduce the image processing time.After that,the SVM classifier is trained twice.The first training is used to sample difficult samples,and the second time the difficult samples are brought into the training set to obtain the final model.(5)Finally,in view of the fact that the SVM classifier can only act on images of the same size as the training sample,and because the size of characters is inconsistent in practical application,this paper adopts the multi-scale detection algorithm based on image pyramid and studies the window fusion algorithm based on the characteristics of multi-scale detection.The improved hop-lbg detector and the improved HOG-LBP detector are used to verify the effectiveness of the improved HOG-LBP detector.Through experimental verification,it is confirmed that the algorithm proposed in this paper has certain advantages in pedestrian detection accuracy and single image processing time.