Gait Recognition Based On Adaptive Background Model

The gait is a person's body traveling position when walking. It has the unique advantages, such as recognize people far away, the non- contact , also, it is not easy to hide, and easy to gather. Gait recognition is a new technology of biometrics authentication which absorbs more and more researchers to concern it in recent years, it recognizes people by their way of walking. It has widespread application prospect in the virtual reality, video monitoring, gait analysis of medicine.Gait recognition analysis and handling aim at the gait sequence, so a gait-recognition system usually includes gathering the gait data, gait detection, feature extract, and recognition. In view of the gait recognition technology has the important theory value and practical significance, on the basis of previous studies, Centered on the theme of gait recognition, this article mainly research on several following aspects:1. This paper proposes a new method of detecting the gait based on an adaptive background model, establishes Gauss mixture background model for each pixel of the gait video sequence's each frame; renew the background model real-time according to whether the current frame's pixels can match with the background model or not, and detects the movement region, compared the method of gait detection based on median value background model and adaptive background model, the experiments proved the latter was perfect when handling 0°view of gait sequence.2. In this paper gait representation methods are referred. We use the width vector of human silhouette's outline as the space gait feature, and normalizes the width vector for each frame to smooth over the influence caused by different points on the outline; and uses the width vector in a gait cycle(40 essential frames) of each gait sequence as the time domain feature, and reduces the vector's dimension as far as possible, converts the feature matrix using the PCA method.3. We discuss the gait recognition with different views, train and predict the width vectors on five different views(0o,18°,36o,90o,126°)using the support vector machine, to test the effect on variational views.Finally we realized the above method on the Matlab platform, the experiment result indicated this algorithm could easily be understood and realized, and got the inspiring recognition performance.