Research On Dynamic Pedestrian Detection System Based On Machine Vision Technology

With the rapid development of computer technology and information technology,computer vision technology has been widely used.The pedestrian detection based on video image is one of the important research topics in the field of computer vision.It has important research significance in the field of security monitoring,especially in the academic and industrial research.Since the HOG feature has been proposed for pedestrian detection,many methods have been derived in recent years,all of which have achieved good results.However,there are many difficulties in detection methods,for example how to build more efficient classifier,and how to effectively detect the movement of pedestrians area.This paper analyzes the current development and research status of pedestrian detection at home and abroad,and studies the detection technology involved in pedestrians.In this paper,the classifier model involved in pedestrian detection is studied,analyzed and optimized.At the same time,the moving object extraction algorithm is improved.The main contents of this paper are as follows.First of all,It introduces the basic knowledge of image processing,and three kinds of commonly used features in pedestrian detection are studied and analyzed.The difficulty of pedestrian detection is that pedestrians are vulnerable to occlusion,which will affect the accuracy of pedestrian detection.For whole pedestrians,human heads are less susceptible to occlusion than other parts.In this paper,the pedestrian heads and shoulders as detection objects.After comparing the three features,the HOG feature which can describe the shape of the head of the human body is used as the characteristic of the classifier in this paper.Secondly,the classifier model used in pedestrian detection is studied and analyzed,and the detection performance is verified.In the study of SVM,the performance of SVM with different parameters is tested,and the parameters with better results are determined.In the study of BP neural network,the model with good result is selected,through comparing the performance of the model with different structures.Finally,BP neural network based on Adaboost and BP neural network based on genetic algorithm are studied respectively,and their performances are compared.After that,the BP neural network with the initial weight optimized by genetic algorithm is enhanced,then its performance is tested.After testing,we can see that the performance of the optimized classifier has been improved.Next,in the study of the moving target detection algorithm,the foreground detection algorithm is optimized.After the optimized algorithm,the anti-interference ability is obviously enhanced.Finally,the improved foreground detection algorithm is used to extract the moving object,then the optimized classifier is used for pedestrian recognition.