Gait Recognition Using Angle Statistical Features

Gait recognition is a new biometrics recognition method, which is the process of identifying individuals by their walking manners. Compared to other biometrics recognition methods, gait recognition has an very important advantage, that can be observed at a distance.Generally, gait recognition consists of three parts:gait detection, gait representation and gait identification.The author attempts to study all parts of gait recognition, and presented a new method, that using angle statistical features to recognising gait.The author researched many pre-processing algorithms thoroughly in gait recognition, and set up the background model by using the method of Least Median of Squares to meet the shortfall of needing too much frames by using mean method. Indirect Difference method made it easier to choose a threshold while getting the binary image. Mirroring image made it comparable between the images in different direction.Angle feature extraction provides a new way for gait recognition. Using statistical methods to calculate the angle interval mean and variance of pixels to centroid distance of the normalized gait silhouette images, and construct them as a feature vector. Compared with the method of boundary feature extraction of gait silhouette images, this method has the advantages of simpler algorithm, quicker computing speed, and no need to create complex mathematical models and so on.Feature selection results are compared between two random search algorithms, genetic algorithm and simulated annealing, and finally chose genetic algorithm as the feature selection method for its less iterations and better feature selection results.K-neighbor classifier, BP neural network classifier and SVM classifier are used for gait identification respectively, and all of them obtain good identification results.A lot of experiments are made by the platform of matlab7.5, using the CASIA database provided by Institute of Automation, Chinese Academy of Sciences, and the experimental results demonstrated that the method has good recognition performance.