Research On The Detection And Tracking Of Human Based On Video Image

This paper mainly studies the research of human detection and tracking based on video image.This technology can be applied to intelligent video surveillance,human action analysis,human-machine interaction,video retrieval based on content and intelligent driving system.Although human detection and tracking technology based on video image has experienced rapid development for several decades,it is difficult to find a fast and accurate system to detect the human body and track the target of the moving body quickly and accurately because of the diversity of the human body,the limited resolution of the camera,the complexity of the application environment and the fast demand for video image processing.3 aspects of the contents are listed as below.This paper analyzes and studies the main algorithms of moving target detection.The background subtraction method is used as the moving object detection method in this paper.Analyzing the effectiveness of the background modeling method,focusing on the study of ViBe algorithm,and aiming at the randomness of the update of the background model of the ViBe algorithm,this paper adopts the measures that updating the background pixels in the background model.In view of the ghost and the static target of the ViBe algorithm,this paper uses the variance to distinguish the ghost and the staticl target.For ghosts,the pixel values of the ghost background model are replaced by the corresponding pixels in the next frame.For the static target,the pixel value of the region is taken to deal with the target.When the target leaves the static position,new pixels in the region are used to replace the background model.This paper analyzes three kinds of common detection and recognition algorithms based on classifier.SVM classification algorithm based on HOG feature is used for human target detection.Aiming at the problem of OpenCV's own classifier,the high false detection rate and high leakage detection rate,this paper cuts the positive and negative samples to train the SVM classifier based on the HOG feature.In view of the false detection of the non moving target as the human target,this paper uses the method of combining the SVM classification recognition based on the HOG feature and the improved ViBe algorithm.In respect of moving human body tracking,this paper proposes a hybrid particle filtering algorithm based on detection.The important density function combines SVM classification detection.SVM can effectively detect the target entry scene and disappear,leaving the scene.In order to achieve the complementarity of information between features,the color feature and texture feature are fused in this paper in order to use the inadequacy of color feature or texture feature.For multi-target tracking,this paper introduces hybrid particle filtering algorithm,and a single component can realize the sampling process autonomously,which is helpful to improve the problem of the reduction of particle diversity because of the resampling of the standard particle filter algorithm.For occlusion problem,this paper adopts a way to update the target state without updating the target model.