Gait Recognition Research And Implement Based On Image Transformation

Gait recognition's purpose is to recognize the ID of people by individual's walking style. Gait recognition has its advantages which are long distant recognition, noninvasive and difficult to conceal. These features have received growing interest within the computer vision community recently.Firstly, an improved single Gaussian model method to reconstruction the background image and the background subtraction method to extract moving body area are used. The noise of gait image is eliminated by color information shadow elimination method, mathematical morphology method and connected component analysis. The gait period is determined by the period of motion area contour-rectangle's width-height ratio. The novel conception of Gait energy enhanced images (GEEI) is presented. This image is created by superposition every frame which multiply its moments, in the half gait period.Then, a novel gait recognition algorithm based on Geometric transformation is proposed. This algorithm put Geometric transformation on GEEI to extract eigenvector which is used to classify and recognition by NN classifier. A novel gait recognition algorithm based on Curvelet transformation is proposed. This algorithm put Curvelet transformation on GEEI to extract eigenvector. To high-dimensional eigenvector, the Krawtchouk moments is calculated as an eigenvalue to classify and recognition by NN classifier.Finally, a simple gait recognition system is designed and implemented. The experiments are successful and get high recognition ratios, which illuminate that these two methods presented in this paper are effective.