The Pedestrian Detection And Tracking Based On The Video

The method of pedestrian detection and tracking is an important research area in the field of computer vision,which is a multidisciplinary cross area and has a wide range of applications.Given the importance of the pedestrian detection and tracking,the researchers have also proposed many classic methods of the pedestrian detection and tracking.Most of the pedestrian detection methods focus on the integral model method,which are to scan the whole image by using a sliding window to obtain the global features of the image.Although the integral method has a good overall understanding,there is no way to overcome the problem of part occlusion.In the light of the problem about the partial sheltering,this thesis lays emphasis on the method of pedestrian detection using the component method to deal with the problem of the partial sheltering.In addition,most of the pedestrian tracking methods depend on the prior knowledge of the target.While these methods are good for obtaining target information,a single feature or target model does not fit well enough to meet the requirement of tracking different targets.In this case,the thesis focuses on the ptical flow that does not rely on the prior knowledge of the target and uses the kalman filter as a search algorithm to achieve good tracking of pedestrians.Based on the study of the traditional pedestrian detection and tracking method,this thesis analyzes the common difficulties of pedestrian detection and tracking and completes the following major research efforts:In terms of pedestrian detection,this thesis puts forward a method of the pedestrian detection that combines the mixtures-of-parts articulated model with the linear regression model.Firstly,the mixtures-of-parts articulated model divides the body into some parts,such as the head,the torso and the limbs.Then,the body is marked with 26 bounding boxes.Next,the human body can be reconstructed according to the spatial relationship and the relevant constraints.Finally,the data of the component bounding boxes is input into the linear regression model to generate the final bounding box that contains the whole human body.The experimental results show that the method of this thesis not only overcomes the problem of the partial occlusion,but also has a good detection effect.In the case of the pedestrian tracking,this thesis presents a method of tracking that combines the optical flow with the Kalman filter.Firstly,the method of optical flow is used to deal with the video frames.Secondly,the information of the moving target can be obtained accurately through the processing of the optical flow clustering and the improved median filter.Finally,according to the information obtained from the target,the kalman filter is implemented to make the predictions of the moving object and achieve the effective tracking.The effectiveness of this method is also verified by the results of experiments.