Pedestrian Detection Based On Improved HOG And Combined Kernel Function SVM

To analyze images and video sequences that have been acquired to determine whether there are pedestrians and give pedestrian's location information.In the field of computer vision,pedestrian detection has a wide range of applications,both in intelligent video surveillance,human behavior analysis,and in vehicle-assisted driving.Analysis and detection of pedestrians has important research significance,it's one of the research hotspots of machine learning.At present,pedestrian detection technology is mainly divided into background-based modeling and statistical learning-based methods.The early research focuses on the background modeling method.Firstly,the foreground area and the background area are effectively segmented,and then the target area is classified.This method is susceptible to environmental and other interference factors.Since 2005,pedestrian learning based on statistical learning has developed rapidly and has become a research hotspot in this field.Based on the statistical learning method,the pedestrian characteristics are first extracted and the classifier is trained.At present,the widely used pedestrian detection framework uses feature gradient histograms for feature extraction and then trains SVM classifiers.This paper is also based on statistical learning methods to achieve pedestrian detection,mainly to do the following work:In this paper,the improved HOG algorithm is used to extract features from the images.The HOG feature vector is extracted in the local cell region to avoid the influence of illumination and geometric changes.In this paper,when calculating the HOG feature vector,the integral graph is used to simplify the calculation.For the HOG feature vector dimension,the extracted features have a lot of redundant information affecting the recognition speed and recognition rate.The improved HOG algorithm is used for feature extraction,the principal component analysis algorithm is used to reduce the dimension of the HOG feature vector,and then start classification work.This paper uses a PSO-based combined kernel function SVM algorithm.In order to balance the globality and locality of the SVM classifier,the classifier has the ability of learning and generalization,and the pedestrian detection classifier with better recognition performance is obtained.The combined kernel function SVM is used to construct the pedestrian detection classification,the linear combination radial basis kernel function and the polynomial kernel function are used as the kernel function of SVM to improve the classification performance.The factors affecting the classification performance of the combined kernel function SVM are analyzed.The polynomial kernel function parameter,the radial basis kernel function parameter,the combined kernel function coefficient and the penalty factor are analyzed.In order to obtain the global optimal solution,the PSO algorithm is used to optimize the SVM parameters of the combined kernel function,and the final pedestrian detection classification model is obtained.On the INRIA dataset,a good pedestrian detection effect is obtained.