Gait Recognition Based On Information Fusion

Gait recognition is a new and emergent research topic in the field of biometric recognition, which makes use of human walking pattern to recognize and identify. It mainly processes and analyses with dynamic image sequences, which has a wide spectrum of promising applications in many domains such as virtual reality, smart surveillance and perceptual user interface.Generally, gait recognition consists of three parts: preprocessing of gait sequences, feature extraction and classification. Feature extraction is the most important part, on which this paper focuses.The preprocessing of gait sequences is extract human motion from gait video. It mainly includes background model, foreground detection and morphological post processing. The segmentation of gait silhouettes has very important influence on feature extraction and subject classification.In the feature extraction step, two effective methods are provided. A new feature abstraction method that is based on Radon transformation is brought forward. According to human's walking characteristic, one period feature vector to represent the whole gait sequence is adopted, which takes object's dynamic and body structure parameters into account. Then, eigenspace transformation based on principal component analysis is applied to reduce the dimensionality of the input feature space. Then, another novel gait recognition system that is based on gait energy image and two directional two dimensional principal component analysis((2D)~2PCA) is put forward. Gait Energy Image(GEI), which is a method to add and average a period gait images into a single image and contains the gait information about frequency, phase, figure and so on, can decrease data and save gait information better. (2D)~2PCA , simultaneously considering the row and column directions, is applied to abstract the feature vectors which has immense contribution for classification.For the gait diversity under different views, the gait sequence which has wrong identification under some one view may has right result under another view. The information fusion based on the Rank rule is adopted, which can effectively improve the gait recognition rate. Considering the limit of single feature extraction method, this paper provides a multi-feature fusion rule which is based on score.The methods are evaluated on CASIA gait dataset which has 124 persons. Results show that the method has encouraging performance; especially small influence can bring in the taking things. The method which is based on GEI and (2D)~2PCA can work well under multi-views, while information fusion even obtains higher recognition accuracy...