Gait Recognition Based On The Realization Of Identity

Gait recognition as a new application in the computer vision, which is more and more of our attention, especially in the countries of the increasing importance of the terrorists, all countries on the security aspects of monitoring and biological identification of a growing concern. As a new research aspect, gait recognition can not be ignored by its advantages.Gait as the only biometric identification technology, which can be remote, is more and more attention to social and scientific research organizations. In social security, identity authentication and video surveillance have a wide range of applications. Especially after the 9.11 incident, Governments who are increasingly concerned about the safety of the community, and hoping to rely on authentication technology to advance warning and counter-terrorism, which effectively put an end to the occurrence of the risk. Compare with other biometric authentication techniques (such as face recognition, iris recognition, fingerprint recognition, etc.) is different from the external human gait, dynamic performance, and in close contact and the time and space. At the same time,compare with the other biometric technologies, which are based on the static characteristics, gait recognition has a non-contact, non-invasive and difficult to hide such a significant advantage. In particular, long-range detection of gait can be achieved (> 5m), and can sense the low-resolution monitor, which have a stronger robustness. Secondly, the gait recognition can be used when the observer doesn't find, and the detection does not need to be co-operation by observers. Third, other authentication techniques can be hidden, for example face recognition in order to cover my face with a hidden, but the gait is more difficult to hide.This article is aimed at the advantages of gait, and researched the relevant aspects, related to the realization of the algorithm. The main work of this paper is a gait recognition algorithm, mainly related to gait recognition preprocessing, feature extraction, data dimensionality reduction, as well as the classification and identification of such a process. Gait recognition of pre-processing stage, this paper introduces the main information on the pre-processing part of gait recognition, mainly involving the use of different scenes in different ways to achieve the context of the process modeling, application of the median method, the smallest value of the number of square method, Gaussian mixture model based on the background by subtract, access to the human contour from the background. In the feature extraction, distance which the contour points are subtracting to contour centroid, as the characteristics. As a result of the data is too large, so the use of data dimensionality reduction to the operation, the paper used in the PCA principal component analysis method to achieve this process, to achieve the purpose of data dimensionality reduction. In the classification and identification is the main way the Mahalanobis distance to measure the distance, and then through the neighborhood recently to get in the way it is kind of gait.In the chapter one of this paper, we introduce the characteristics of gait recognition, and as well as domestic and foreign scholars for research to identify gait and cognitive level, but also introduce some of the mainstream of the current gait recognition algorithm for classification, mainly based on the model and Appearance-based methods. And then introduced a more detailed look popular algorithm are mainly based on physical parameters method, based on the outer contour method, based on the solid outline of the method, based on the Fourier transform method. Finally, this chapter is introduced the gait recognition and realization of the whole process.The second chapter introduces the method of gait identification of preprocess, when a frame come ,which contains the target, we should first of all, the background of this image modeling, and thus extract this target, this chapter is to introduce the contents of this content. For the background of a single, and relatively easy to achieve, we can use the background model is relatively simple, such as the three frames, in value and so on. For the Relatively large changes in the background, the chapter describes the background based on the difference method of clustering by adaptive algorithm. This algorithm can solve the scene which is a more obvious light change. After Background model in the establishment, it is necessary to extract the target in order to obtain this target, here the use of threshold segmentation and background reduction method to get. When we get the target, due to the existence of noise, issues such as the target is not connected, where the use of binary images based on morphological methods to solve the above problems.The third chapter introduces the process of gait recognition on the need to use some algorithms introduced mainly on the feature extraction, data dimensionality reduction and classification algorithms. The beginning is from feature extraction, the basic content of the images described, including with the adjacent neighborhood, the perimeter of the images and so on. In the data dimensionality reduction is mainly introduced the Principal Component Analysis PCA approach. In the distance measure, the main information on the distance is Mahalanobis distance and Euclidean distance. In the classification mainly on the latest neighborhood, K-neighbor method, and support vector machine aspects. These algorithms are in the process of Chapter four are applied to a certain extent, for the realization of gait recognition has been the basis for the theory.The main contents of Chapter four is the whole gait recognition algorithm of the process of gait recognition from the beginning of the preprocess using the LMedS (the smallest value of the number of square method) the method of background modeling, and then the current frame subtract background model getting the target prospects. The prospect of access to binary morphological operation, more complete extraction of the target of the human body blob. Access to the body blob, the contours of the body on the extraction, carried out anti-clockwise tag, and the outline for the distance calculation, getting the total perimeter, and then part it to sub-360, where the use of interpolation methods and then calculating the distance to the centroid of contour.Because of so many distance vector which we calculating from contour to Center of mass, so it dimensionality reduction operation, in this paper used PCA (Principal Component Analysis) to obtain the characteristics of the major components, and then to select the value through the contribution of the main characteristics of the first few components, projection further operation, to obtain projection feature vector. Feature vector and then used the method of Mahalanobis distance to measure the distance in the classification and identification of each gait database to get the cluster center, and then use the method of nearest neighbor classification, belong to the minimum value obtained gait class, if the value is greater than the threshold set, the gait do not belong to any class.The fifth chapter is to introduce the main results and some future prospects.