Pedestrian Counting Based On Real Traffic Scenes

As a general-level technology in the field of artificial intelligence,computer vision is now widely used in transportation,medical care,and security.Pedestrian counting based on computer vision is not only important for smart security,intelligent transportation,but also provides data analysis for new retail and smart cities.Pedestrian counting in different scenarios has different meanings.This article focuses on the pedestrian counting based on fixed monocular cameras in real traffic scenarios,and designs and implements an effective pedestrian counting scheme.In this paper,because of the excessive interference of passing vehicles and shaking trees in real traffic scenes,pedestrian detection is based on static image features.HOG's expression of gradient information makes it perform well in pedestrian detection,but due to the lack of expression ability of HOG in real and complex traffic scenarios,and LBP can express local texture features,so it combines LBP with HOG to detect pedestrian in this paper.At the same time,this paper uses CNN convolutional neural network for pedestrian detection,and linear SVM is selected as the pedestrian recognition algorithm.Taking into account the needs of real traffic scenarios,pedestrian detection is performed in the adaptive region,two virtual count lines are demarcated in the region,and the pedestrian tracking is established based on the Kalman filter.Counting by judging the cross line behavior of pedestrians.The experimental results show that the accuracy of the LBP-HOG in this paper are higher than the LBP feature and the HOG feature in the hybrid data set based on the real traffic scenario during the feature selection and testing phase,and it is similar to CNN feature performance.The linear SVM chosen in this paper enables highly accurate linear classification of test samples.At the pedestrian detection stage,this method has strong anti-occlusion ability,is insensitive to pedestrians' wear and movement,is insensitive to environmental background and illumination,and is less interfered by motion noise,effectively avoids the shortcomings of pedestrian detection method based on motion information is too sensitive to the movement in the environment.In the pedestrian tracking and counting phase,this method can predict and match the position of pedestrians and effectively solve the problem of pedestrians being obstructed or missed during the tracking process.Moreover,the method of area adaptation reduces the search range of pedestrians and reduces the impact of environmental information.The experimental results show that the method of this paper has strong ability in detection and can meet the needs of real traffic scenarios in tracking and counting.