Full - Attitude Attitude Estimation Of Rigid Objects In Vision Assisted Technique Of Visually Impaired

China has a large visually impaired population, which makes the research on vision aid for the visually impaired very important to the country. Getting the all-dimensional pose information of the objects is one of key steps for vision aid. However, there are some problems for the existing methods estimate the all-round posture of the objects. Therefore, we present a method based on manifold structure for measuring the all-round posture of the rigid object.The main contribution of this paper has the following four aspects:1. Design a scheme based on manifold structure for measuring the all-round posture of the rigid object.Firstly, collect the equal angle-interval object images and reduce the dimension of the image data by manifold learning algorithm; nextly, obtain the regression relation model between the posture image and the manifold coordinate by training the prior data; lastly, get the manifold coordinate of the new images to calculate the pose angle of the object.2. Present a new manifold learning algorithm for pose estimation of the object.According to the characteristics of the image acquisition sequence, we optimize the computation part of the geodesic distance matrix in the Isomap algorithm and then present a new Isomap manifold learning algorithm. The new manifold learning algorithm is more than twice as fast as the old manifold.3. Design an algorithm for estimating the manifold dimension.The manifold dimension estimation is one of important procedure for manifold learning. So, we design an algorithm based on the PCA to estimate the dimension of the manifold which guides the dimension reducing of the image data.4. Design a regression model between the image data and the manifold coordinates.Since that there is not an explicit formulation between the image data and the manifold coordinates, we design a regression model which can predict the manifold coordinate of the new posture image.In the last, we verify the effectiveness of the proposed method in this paper by using the 3D model image and the real object image. With the same num of the train images, the result shows that the proposed method is better than the traditional method which is based on the machine learning.