Research On Pedestrian Detection And Tracking Algorithm In Traffic Flow Video

Pedestrian detection and tracking technology in video is a very important research topic of computer vision,widely used in intelligent transportation,Advanced Driver Assistance Systems(ADAS)and other fields.As pedestrians are non-rigid objects,pedestrian gestures and clothes color are complex and changeable,coupled with changes in lighting conditions,pedestrian detection and tracking technology research has become a difficult field.In this paper,the research work involves feature extraction,classification and tracking algorithm in the field of pedestrian detection and tracking technology.We proposed a new pedestrian detection and tracking algorithm based on the extended center-symmetric local binary pattern(XCS-LBP),cascaded AKSVM classifier,genetic algorithm and particle filter tracking algorithm.And we get a better detection tracking effect.The specific work and achievements of this paper are as follows:1.For pedestrian detection.Firstly,the XCS-LBP feature vectors of the training sample set are extracted,which can describe the information in the image and improve the detection precision.Then,weak classifiers are adaptively trained,the cascade order of weak classifier is determined by genetic algorithm,and the final classifier is obtained.Finally,we get the pedestrian detection region.Experiments show that the training classifier has a good classification effect,and ensures a better detection accuracy.2.For pedestrian tracking.Firstly,we collect the video every 5 frames and use the cascade AKSVM classifier after training to find the pedestrian area,and mark the pedestrian with the rectangular box.Then,we track the target pedestrian using the fast particle filter algorithm based on the color histogram the next frame in the video.By comparing the similarity of the color histogram of the particle region and the color histogram of the target region,and we obtain a new position of the target.And the experiment achieved good results.