| In recent years, with the development of the society, there are higher requirements for the recognition of the identity of human being itself in catering for many occasions. Because of many existing deficiencies, traditional methods of identity recognition have been gradually replaced by the new technology of biometric recognition. Owing to its dependence on the characteristics of human being itself which can't be replaced by traditional identification technology, the new biometric identification technology has been researched energetically and extended broadly in safety field. Gait recognition, as the second-generation biometric identification technology, not only overcomes the defects existed in the traditional technology of identity recognition, but also has peculiar advantages of being non-invasive,non-contacted,hare hided and camouflaged,easily collective compared with the first-generation technology of identity recognition, it has been increasingly welcomed by many people.Gait recognition is a new and developing identification technology of biometric features, which aims at identifying the feature of personal identity through the examination of people's walking postures. Early biometric medical technology shows that gait information of mankind is unique, so it is feasible to rely on people's walking postures to identify one's identity features. In this paper, the author mainly made an exploration and study on the extraction of gait features based on moment invariants, recognition and processing in the shadow of gait image which is the main factor influencing the extraction of gait features.In the aspect of image preprocessing. The author used average-time method to establish models. And in the light of real-time requirement she applied dynamic background updating to external environmental changes, examines image motional area through the subtraction of background, and then does the binarization. After that, the author transferred the obtained foreground image into a binary image and finally got the movement contour of human body through a series of morphological analysis.In the aspect of the extraction of gait feature, After the attainment of the movement contour of human body, the author applied conventional Hu moments to the extraction of gait feature, and then found and pointed out the limitations of traditional Hu moment invariants in the side of extraction of gait features. On this basis she improved by doing an extraction of gait features through a new kind of moment invariants. The new moment invariants has been greatly improved in the perspective of computing speed, stability, and the period of gait image as well as overcoming the shortcomings of traditional Hu moment.In the aspect of gait recognition, the author adopted a simple K-nearest neighboring method. The following is how K-nearest neighboring method works: Calculate the Euclidean distance between Tested samples and the known samples, find out K sets of neighbors which is nearest to Euclidean distance, if the majority of k nearest neighbors belongs to a certain category, then put the tested samples into this certain category. The author did an elaborate experiment in the data base of CASIA gait. The concrete procedure is as follows: Choose a certain set of gait sequence as tested sample, three sets of gait sequences selected from each of 20 people as known sample, and then match between the tested sample and the known sample. The results show that the rate of accurate recognition accounts for over 90%.Finally, the author conducted a study on the shadow which is the main factor influencing the extraction of gait image. After comparing the traditional shadow elimination of RGB space with that of HSV space, the author put forward a new method of dealing with shadows. In view of the shadow of human movement and the fringe information of sporting area, this method only depends on the gray image and can efficiently remove most of the shadows with the premise of not affecting the target itself. |
PDF Full Text Request