Research On Multi-Perspective Gait Recognition Algorithm Based On Bemd

As an emerging biometric recognition technology,gait recognition can determine the identity of pedestrians based on the walking attitude of pedestrians.Compared with the traditional biometrics,it has the advantages of non-invasive,setting easily in public areas and hiding difficultly.At present,gait recognition research mainly focuses on the specified angle,which is difficult for the current gait recognition aiming to multiple perspectives.For the gait recognition system is often affected by the angle of the observation,it is very important to research the multi-perspective gait identification.This thesis presents a multi-perspective gait recognition algorithm based on BEMD.According to CASIA database provided by Chinese Academy of Sciences.The gait image pre-processing,gait feature extraction and gait recognition are studied detailedly.The main contents are content is as follows:In the pre-processing part of gait image,the binary image of the gait contour is obtained by extracting the contour of the moving target from the background image using the background subtraction method.Some noises and small connected regions are removed by using the operation of morphological processing and connected region processing.After the normalization of the image,the pedestrian profile is adjusted to the appropriate size,and on this basis,the gait cycle can be detected.In the gait feature extraction part,gait energy image is obtained,in a gait cycle and the bidimensional empirical mode decomposition(BEMD)method is used to decompose the gait energy image,and the first three high-frequency components of the decomposed four components are reconstructed and regarded as the gait features.The gait features of each perspective are added,and the multi-perspective gait features are finally constructed.Analyzing and comparing the accuracy of different angles when each modal component and combined components respectively regarded as the features,the experimental results show that the classification effectiveness of regarding the first three components as the features is better than the other features.Moreover,comparing the average recognition rate of each angle with the recognition of the multi-perspective gait recognition after the angle fusion,and it can be proved that the multi-perspective gait feature has a good classification effect.Gait recognition part,the constructed multi-perspective gait feature is classified using support vector machine(SVM),the penalty parameters and kernel parameters of the model are optimized using genetic algorithm(GA).Using the public CASIA database,the average recognition rate can reach 93.52%,and higher than the recognition rate 90.22% before optimization.