Research On Pedestrian Detection Algorithm In Video Surveillance

Pedestrian detection in video surveillance has always been a hot and key research topic in the field of computer vision,and the research of pedestrian detection is of great significance to various aspects of human development.Now,with the development of artificial intelligence and the popularization of intelligent video surveillance system,the application of pedestrian detection in video surveillance will be more and more extensive.It not only can be applied in people's life,but also plays a very important role in agriculture,navigation,aerospace,military,the national defense security and other important fields.In pedestrian detection,how to get the most accurate detection results has always been a key issue that pedestrian detection technology researches.To address this issue,the main research work in this paper is as follows:This paper summarizes the research background and significance of pedestrian detection,and the research status of pedestrian detection technology at home and abroad.This paper introduces the common feature extraction algorithms in pedestrian detection,and analyzes their principles and implementation steps.The Histogram of Oriented Gradient(HOG)algorithm is mainly introduced.The feature principle of HOG algorithm is to conduct statistics on the histogram of gradient direction in the local area of the image,and making the content of statistics into features.Then,this paper introduces the common classification algorithms.The Support Vector Machine(SVM)algorithm is mainly introduced,SVM algorithm is a classifier based on the idea of optimal classification surface.This paper introduces superpixel,superpixel classification and Simple Linear Iterative Clustering(SLIC)algorithm.Superpixels represent a collection of pixels with similar characteristics,and it can be generally divided into the method based on graph theory and the method based on gradient descent.SLIC algorithm is a superpixel generation method based on gradient descent.This paper proposes a pedestrian detection algorithm based on superpixel and HOG.The purpose of the algorithm is to improve the accuracy of pedestrian detection and the accuracy of recognition.Before feature extraction,SLIC algorithm is used to preprocess the image,then,HOG algorithm isused to extract features from the image.SVM classification algorithm is used to classify features after feature extraction.In the experiment,the weight factor w value of the set color and spatial difference is 50,and the number of superpixels k value of the same scale of pre-segmentation is 500.When HOG algorithm is used to extract features,an8×8 pixel rectangular region is used as the cell for HOG feature collection,and each cell does not overlap.Finally,it is sent to the SVM classifier for training.Using this algorithm,INRIA pedestrian database and MIT pedestrian database were trained and tested respectively.In this paper,while ensuring the robustness of HOG algorithm,miss rate and error detection is reduced,and precision,recall and F1-score is enhanced.Experimental results show that the algorithm has good detection effect on pedestrian detection in video surveillance.This paper introduces Principal Component Analysis(PCA)algorithm.PCA algorithm is a commonly used method based on variable covariance matrix.Since there is a lot of redundant information in HOG feature,PCA algorithm can be used to reduce the feature dimension.In the experiment,the average value of the HOG feature of all corresponding pixel points in the training sample was calculated.According to the HOG feature vector of the training sample and mean value of the training sample,PCA algorithm is used to calculate the feature value,feature vector and covariance matrix.Then,select the first P principal components of the calculated covariance matrix,and feature dimensionality reduction is conducted for each HOG feature value in the training sample.In this way,the HOG-PCA feature is obtained,and the new p dimensional HOG-PCA feature is sent to the SVM classifier for training.Using this algorithm,INRIA pedestrian database was trained and tested.Experimental results show that the HOG-PCA feature algorithm can not only reduce the training time of pedestrian recognition in video surveillance,but also greatly improve the recognition rate.