ABSTRACT
Human beings can form perceptions about their 3-D world through many years of their experience. Building perception for machines to recognize their 3-D world by visual information is difficult, as the images obtained by cameras can only represent 2-D information. This chapter demonstrates one approach for understanding the 3-D world from 2-D images of a scene by Kalman filtering. Before employing Kalman filters, the images, however, have to be pre-processed and segmented into objects of interest. The chapter thus begins with low and medium level image processing that deals with pre-processing and segmentation of images. Next the principles of perspective projection geometry are covered to explain many interesting phenomena of the 3-D world. The recognition of images with self-organizing feature maps and the principal component analysis have then been taken up as case studies. The chapter ends with reconstructing the 3-D world from multiple 2-D images by using Kalman filtering.
