Pedestrian Detection Based On HOG Feature And Multi-Classifier Ensemble Learning Method

With the prosperity of artificial intelligence technology and the arrival of the information big data era,computer vision technology has become an indispensable part of the AI field.At the same time,pedestrian detection technology as a hot issue in the field of computer vision,has also received more and more attention from relevant scholars.Pedestrian detection technology is an image analysis technique that determines whether there are pedestrians in the image and accurately locates the pedestrians appearing in the image from the background.This technology has played a key role in engineering applications,academic research and smart technology practice,and has received extensive attention from people in related fields.After years of development,pedestrian detection technology has achieved certain achievements.Nowadays,this technology has been widely used in the fields of intelligent monitoring,airport security,vehicle assisted driving and intelligent robots,and has become a research hotspot in the field of target detection.At present,there are two main research methods for pedestrian detection.One is based on motion detection algorithm.This kind of method firstly extracts the foreground of the movement of the target to be detected by using background modeling algorithm,then classifies it by classification algorithm,and identifies whether there are pedestrians through classification results.This method is simple and fast,but it can only detect moving targets and is vulnerable to the influence of light,shadow and weather environment.The other is based on machine learning.Human body has its own appearance characteristics in both real life and image.Machine learning algorithm is good at automatically learning different changes of human body from a large number of sample data sets.Therefore,we can automatically build human body model by selecting the relevant human body features in the learning samples of machine learning algorithm.This method has strong robustness to different samples.At the same time,it can avoid many unfavorable conditions by selecting training samples artificially and selecting reasonable classifiers.Based on the advantages of the second method,this paper uses the HOG feature proposed by Dalal as the describing operator of human body features.The traditional pedestrian detection algorithm uses a single classifier with low recognition rate,which is easy to cause false detection.The problem is to propose an integrated learning method using multi-classifier integration to perform pedestrian detection on images.Firstly,the gradient direction histogram(HOG)feature is used to extract the feature vectors needed for training different classifiers.Then three component classifiers,Naive Bayesian(NB),Logistic Regression(LR)and Support Vector Machine(SVM),are trained by using the extracted features.Then,the three component classifiers are integrated by weighted voting strategy.Finally,a strong classifier is formed and a pedestrian detection method with higher recognition rate is obtained.Finally,the experimental results using the proposed method and various traditional detection methods on different pedestrian detection datasets show that the proposed method has higher detection accuracy and reduces false detection and missed detection.Overall,It has better detection effect and improves the robustness of pedestrian detection and recognition.