Research On Multi-scale Pedestrian Detection And Tracking Method Based On Integral Channel

In recent years,cars have gradually become vehicles that people rely on for their outing.With the significant progress made in the field of artificial intelligence,intelligent assisted driving systems have become more and more hot.Pedestrian detection technology is an indispensable part of car-assisted driving.It can effectively and accurately identify pedestrians in front of vehicles,which can effectively reduce the occurrence of traffic accidents.In practical applications,due to the open environment,considering the factors such as pedestrians' clothing,posture,and occlusion,it is also necessary to consider the influences of road conditions and light changes,making it perfect for pedestrian detection technology.Therefore,improving the accuracy,real-time and robustness of the pedestrian detection algorithm is a key issue for practical application,and has far-reaching research significance.This paper is based on the “National Key R&D Program”(2016YFB0101102),“Electric Vehicle Intelligent Assisted Driving Technology R&D and Industrialization”,with the research of vehicles and vehicle cameras under outdoor driving conditions,mainly researching how to pass visual inspection and tracking.The algorithm realizes that the vehicle can effectively detect and track the pedestrians who influence the driving in various sports modes.Pedestrian detection and tracking tasks in the driving process are realized by the pedestrian detection and tracking system.This paper has completed the following four tasks through a large number of literatures and experiments:1)The research background,significance and research status of pedestrian detection and tracking technology at home and abroad are introduced in detail.The pedestrian characteristics were further studied,and the HOG features and integral channel characteristics were selected.After subsequent experimental analysis,the characteristics of the integrated channel were used as the pedestrian detection characteristics,and the pedestrian feature extraction experiment was completed.2)Classifier selection.The traditional pedestrian detection method HOG+SVM is analyzed and found to be insensitive to pedestrians with high foreground and background similarity.Due to the complex external environment,single-feature pedestrian detection can not meet the demand,and the multi-feature detection pedestrian algorithm is further studied.The categorization classifier Adaboost is selected as the classifier.Firstly,Adaboost is trained,and finally the trained Adaboost classifier is used as the pedestrian detection classifier.3)Research on image fastness algorithm.In order to improve the real-time performance of the pedestrian detection algorithm,the image pyramid algorithm is improved,and the characteristics of the integral channel at different scales are estimated.Then the features are classified by Adaboost classifier.The experimental results show that the proposed algorithm can improve the accuracy,rapidity and robustness of pedestrian detection,and provides a stable pedestrian detection algorithm for the assisted driving system.4)Pedestrian tracking.To ensure the security of the system,a tracking algorithm is added.The Mean Shift algorithm(Mean Shift),adaptive Mean Shift algorithm(Cam Shift)and particle filter algorithm are studied.Considering various external factors,the tracking algorithm is further studied.Finally,the Kalman algorithm is selected for tracking.excellent results.Finally,the proposed scheme uses VS2013 and Open CV2.4.9 simulation experimental system for experimental research.It realizes the effective detection and tracking of pedestrians during the driving process under different working conditions.