Gait Recognition Based On Gait Optical Flow Graph And Sparse Representation

With the development of science and information technology, security has become a globalconcern for all people. How to ensure the safety of people and identify unknown people effec-tively has become our common concern. In order to solve those problems, we propose a biomet-ric identification technology. For example, fingerprint, iris, face and so on. These biologicaltechnology often need identified objects, which requires even need to be in close con-tact.However, It is often not know, non-contact in the security monitoring field to be identifiedobject. The gait recognition can be used in the long distanceand non contact.identified in the longdistanceand non contact case. As a new biometrics,Gait recognition has Practical and researchvalue in the area of information security and public security. In this paper, our main contributionsabout gait recognition are as follows:1. Research on the gait detection algorithm. Firstly, the image need background modelingwith the median method.Then, Several gait detection algorithms are compared. The calculation ofoptical flow method is too complicate and require efficiently computer.Frame difference methodis easy to produce cavitation. But our database background is relatively simple, using the back-ground subtraction method to detect. At the same time,the imge is removed the noise of motionregion and single hole by morphological processing and eight connected region analysis.Last, Itget gait cycle through the ratio curve of human gait in image sequences.2. Based on optical flow, this paper propose a gait optical flow graph algorithm.At first， thegait image sequences use Horn-Schunck algorithm for synthetic optical flow. Then,the gait opti-cal flows are Superposition averaged to be a gait optical flow graph.But the dimension of gaitoptical flow diagram is high, easy to cause the curse of dimensionality. At last, with compare al-gorithms，the flow graph is reduce by PCAand LDAtogether.3. Design the sparse representation classifier at last. Due to the sparse representation can begood at classify the image,and used for gait recognition as well. On this basis, this paper proposean improved sparse representation for gait recognition that is the locally constrained.At first,Thispaper analysis the advantages and disadvantages of group sparse representation and locally con-strained sparse representation. Considering the advantage of the above two kind of sparse repre-sentation, constructing a model of the locally constrained group sparse representation, make thesame category sample structure into a group, and mark it with local constraints. Then using thesubgradient optimization algorithm, design the locally constraint group sparse representation forclassifier.4. Using CASIA Dataset B gait database of the Chinese Academy of Sciences simulate theabove algorithm.At first,we select the part of human gait video data for gait detection and gaitoptical flow calculation, and get obtained gait feature vector by reduce. At last, training the local group sparse for classifier, The experimental results show that, The algorithm has an adcantageover the classical algorithm, has better recognition rate.