An Extension of Locally Linear Embedding for Pose Estimation of 3D Object

Diverse pose estimation of 3D object in the whole view-space is a problem perplexed many researchers. In this paper we propose an algorithm extended from LLE which can estimate the arbitrary pose of 3D object in the whole view space. First, we compute the eigen-images of training set by introducing the idea of PCA using the low-dimensional embedding coordinate deduced from LLE. For a new sample we can compute its projection to the eigen-images, and the nearest training images from the new sample are the estimation poses. Next, we set different weight for different projection direction depends on its eigen-value when computing the distance between the new sample and the training images. Experimental results obtained demonstrated that the performance of the proposed method could estimate the diverse pose of 3D object efficiently and precisely, also our algorithm can be extended to real-time pose estimate, is of a potential future.