Object pose detection in the presence of background clutter and occlusion

This work explores image processing techniques that involve the application of eigenspace methods for pose detection. An eigenspace method for data compression used in the image processing field is commonly referred to as principal component analysis (PCA). We present some recently introduced eigenspace concepts for detecting the pose angle of an occluded object located in an image containing background clutter. To detect the pose of a target object in the presence of background and occlusions we analyze two eigendecomposition methods. The quadtree structure includes dividing the training images into quadrants and creating a subspace eigendecomposition for each level. A statistical robust approach is also applied that weights the background and occlusion pixels based on their influence on the reconstruction of the desired target object. We review both of these pose detection approaches and illustrate each application with an example.