Research On Pedestrian Trail Tracking Based On Gait Analysis

With the increasing attention to public security,video surveillance has been a part of people's life.Intelligent monitoring needs have become imminent.Used to find the target pedestrians from the video are more and more widely,which is by tracking problem has become a hot research topic.The biometric technology based on gait analysis emerging,is through the pedestrian walking posture for human identity verification rely on its unique advantages has become one of the research focus.First of all,based on the traditional minimum mean variance method of background subtraction,in order to overcome the problem of hollow in the minimum mean variance and reduce the computational complexity,a multi-threshold background modeling with weights was established to establish a dynamic background model.Then,the moving objects in the foreground are filtered by median filter to remove the noise.The human detection method based on HOG model is used to detect the pedestrians.Morphological processing is performed on the outline of the human body initially to obtain a clear body contour.Secondly,Secondly,gait feature extraction method is improved,it can extract gait energy image histogram of oriented gradient(HOG)descriptors,the local information entropy feature descriptor and Potter descriptor,which can represent gait features more fully embodies the local characteristics of gait,gait energy image and edge frequency gait characteristics of gait feature.After that,the simulation experiment shows that the improved gait energy map feature representation method has a higher matching rate.The PCA algorithm and the SLLE algorithm in manifold learning are used to reduce the dimension of gait features,and the inherent structure of the data is determined.It not only eliminates the shortcomings that the linear PCA algorithm can not show the gait feature structure,but also overcomes the problem that the SLLE dimensionality reduction algorithm has a high time cost.Experiments show the effectiveness of fusion PCA and SLLE algorithm in recognition rate.Finally,the article gives the fusion of gait features and pedestrians' static features to trace the human tracks.Considering the application of Hu invariant gray-scale images,in order to enhance the static characterization of pedestrian appearance and clothing color,the pedestrian images YUV(Y,U,V),RGB(R,G,B)and HSV(H,S)Eight-color channel Hu invariant,so as to ensure that there is no significant increase in the dimension,but also get the ideal static characteristics.Thus a gait tracking framework based on fusion of gait features and dynamic and static features is obtained.Visual Studio 2015 is used as the development environment and the OpenCV,a computer vision open source library,is used to design and implement a human tracking system based on the fusion of gait features and static characteristics of pedestrians based on the MFC dialog.The system can sort the matching results of the target pedestrians and give the top five matching pedestrians matching the target pedestrians to ensure the real-time and accuracy of tracking.