Human Object Detection And Tracking Based On Gait Recognition

With the development of computer technology,intelligent video surveillance technology has been widely used in many fields,such as civil,military,aviation,medical and human-computer interaction,and its importance is becoming more and more prominent.Moving object tracking is an important branch in the field of computer vision.At present,although the object tracking technology based on video has help people solve a lot of problems,but there are still some deficiencies.It is a very important and worthy subject to improve the detection and tracking efficiency of video surveillance.Gait recognition is a kind of long distance identification method,which is nonaggression,difficult to hide and difficult to camouflage.As a new biometric feature,gait recognition technology has an important research significance and wide application prospect.In the thesis,we use gait recognition to realize identity identification and then to locate and track.It has important applications and significance in security,criminal detection and other fields.In this thesis,for the video sequences with fixed camera,we firstly detect the object region of interest,and then the object recognition is performed by extracting the gait features of the object.The last step is to choose target and track.The thesis involves three main parts:object detection,human gait recognition and object tracking.The main work of this thesis is summarized as follows:(1)In the aspect of object detection,we study and compare the classical object detection algorithms.Then in this thesis we propose an improved method called PVibe which combine Vibe method with the human body target detection characteristics of human body.The experimental comparison of human target detection is in the simple and complex environment.The experimental results show that the algorithm compared with other detection methods,has a good ability to adapt to the dynamic environment and to distinguish between non-human body.Our method has a fast detection speed,can extract the human contour better and has a better detection effect.(2)In the aspect of gait recognition,we first use the object detection and segmentation algorithm to obtain the gait silhouette from gait sequence,and track the body to get the object contour.Then we calculate the distance between contour centroid and each contour boundary pixel.By this way,the distance signals are used as gait feature.BP neural network classification algorithm is used to train the experimental data and to identify the object.We compare the gait recognition rate at different angle of gait sequence.The experimental results show that the highest recognition rate can reach 88.33%.The recognition rate is significantly improved compared with the k-nearest neighbor classification algorithm,which proves the high effectiveness of the proposed algorithm.(3)In the aspect of target tracking,after having achieve the gait recognition,we select the object which we are interested in.And then the tracking is carried out.In this thesis,we study the several mainstream tracking algorithms.And we investigate and analyze the advantages and disadvantages of the optical flow tracking algorithm.Then we propose an improved method called pyramid LK-MVE which combines the optical flow tracking algorithm with motion estimation for tracking the target.Compared with the mainstream of the target tracking algorithms,for similar color and shading conditions,our method achieves a good tracking effect.In the tracking speed,the improved algorithm has significantly been improved.