Research On New Techniques Of Infrared Gait Recognition Based On Information Fusion

As a new biometrics recognition technique, infrared gait recognition is of great application value and theoretical significance. However, the current research in this field is in its "infancy". Efficient gait features and classifiers are badly needed to find out to increase the total recognition rate. Moving body segmentation, gait feature extraction and gait recognition method are the three key stages that affect the total recognition rate. Therefore, this thesis focus on the three aspects mentioned above.Information fusion coordinates, optimizes and blends the information from multiple-sensor and multiple-sources to generate new valuable information to make more accurate and credible conclusion. Information fusion has achieved good effects in the field of other biometrics recognition. Therefore, based on analyzing the basic theory and technique of information fusion, this thesis uses "information fusion technique" as the research approach, makes fusion study on infrared gait recognition in dimensions of data level, feature level and decision-making level and so on. It proposes a new efficient infrared gait recognition technique to solve the problem of the low total recognition rate.In the aspect of moving body segmentation, this thesis analyzes the mechanism of infrared thermal imaging and compares the infrared image with the visible light image. On the basis of the feature of infrared gait image, a new algorithm of infrared gait based on Curvelet transform pixel level image fusion is proposed. Firstly, a Gaussian mixture model background of infrared gait series is constructed, and image difference operation is done on the current frame and background, gets Image 1; secondly SUSAN operation is used to make edge detection on the background image and the current image, and difference operation is done on the edge image of the current frame and background edge image, get Image 2; thirdly, uses the powerful edge direction operation capacity of Curvelet transform to make image fusion based on Curvelet transform with Image 1 and Image 2, then the two images are blended into one rough segmentation image; lastly, the sync pulse release characteristics of pulse coupled neural network (PCNN) are made good use to make fine segmentation and binary processing, and self-adaptive segmentation effect has been achieved by introducing the multiple model immune evolutionary algorithm to determine the optimal segmentation parameters of PCNN quickly. Comparing and analyzing the experimental data, it is discovered that pixel grade image fusion based on Curvelet transform greatly facilitates the optimization of the segmentation effect because the detailed edge information is efficiently blended.In the aspect of gait feature extraction, this thesis analyzes the common gait features and makes good use of the directional properties of Brushlet transform to construct a Brushlet transform complex eigenvalues based on the combination of energy characteristics and phase distribution characteristic as infrared gait new features. Experiments prove that the new feature is efficient. Infrared gait feature fusion refers to constructing fusion features by extracting the gait feature from all the information sources and giving integrated processing to improve the accuracy for further recognition. Thus, analyses are given to the support vector machine and rough set technique as a basis; the advantages of the two techniques are made good use and combined to construct a new infrared gait feature fusion model based on the support vector machine and rough set technique, the determining of its parameters is a convex quadratic optimal problem, and the global optimal solution is to be got. This model makes feature vectors links in the input space and takes advantage of the function of rough set data reduction, so the curse of dimensionality problem has been efficiently solved, and optimal feature fusion has been achieved. Experiments indicate that the new model can efficiently blend various the infrared gait features.In the aspect of infrared gait recognition, this thesis firstly carries out a research on constructing a new individual classifiers recognition model, in which the Contourlet packet is optimized by artificial fish swarm algorithm and a new Contourlet packet recognition model is constructed based on combining the Contourlet packet and redial basis function neural networks; secondly, the biomimetic pattern recognition theory is used to construct a new biomimetic network recognition model on the basis of quantum neural network; thirdly, a study is carried out on the new multiple-sensor fusion recognition technique on measurement-level and decision-level; a multiple classifier fusion technique based on improving particle swarm optimization BP fuzzy network is proposed to make efficient measurement level multiple classifier fusion recognition; a new multiple classifier fusion recognition technique based on quantum genetic algorithm optimization evidence theory is proposed to make efficient decision-level multiple classifier fusion recognition. Simulation experiments show that the two individual classifiers recognition model have achieved good effect, however, the two multiple classifiers fusion recognition new technique has reach the ideal level.The thesis makes a conclusion on the innovation of this research and main achievements. It also puts forward suggestions on the further research.