A Study Of Gait Active Energy Based On Space Decomposition And Joint Sparse Model

Gait recognition as a new biometric technology has been the most popular topic in recent years, which has a special advantage in classification recognition under remote and low resolution conditions. But gait recognition still trapped in the theoretic exploratory stage, far from practical application requirements. How to design and implement a set of gait recognition algorithms which has a high recognition rate, robust, and scalable is key to gait recognition.Based on previous studies, the paper focused on gait active energy image, the research results are as follows:Firstly, the background and significance of gait recognition are introduced, the characteristics, steps and current challenges of gait recognition have been analyzed. Inspired by the current status of research and development, the paper focused on several typical gait recognition method from a different point of view of the advantages and disadvantages for the method. Finally, the research point of the paper is based on class energy diagram.Secondly, existing class energy diagram is unable to eliminate static interference area, the spatial decomposition method was proposed for gait active energy image. Extracting row and column quality vector, optimizing feature vectors through spectral analysis, transforming the problem from time-domain into frequency domain, gait feature vectors are built by adopting extreme value method. Dynamic spatial decomposition method was used to classify gait features in the spatial domain. Adopting the proposed method, gait energy image based gait recognition performed a promising result without reducing feature dimensions.Finally, based on the spatial decomposition of energy image, the paper focused on how to eliminate noise from carryings. With distributed compressed sensing principle in physics,applying the joint sparse model, gait energy image features was divided into common features and private features. In the training stage, comparing feature sequences of samples and the tests,obtaining private gait characteristics to eliminate the interference of carryings. Using random projection of gait features to reduce dimension, the gait feature was reconstructed by minimizingthe first norm, the minimum residual was computed to achieve the purpose of classification.And through other existing methods for comparative experiments, the feasibility of the proposed method was verified. Eventually, under the unstructured characterization of gait feature extraction in this paper, the recognition rate can reach 83%, and have better and robust result under the condition of clothes and bags.