Motion and structure from image sequences (Q1320410)

This book has grown out of our research activities in the area of deriving geometric and motion description of dynamic scenes from visual images. It should prove useful for researchers who are interested in a comprehensive treatment of this topic and practitioners who wish to find algorithms they can use. The results presented here are mostly very recent, some of them being published here for the first time. A rich collection of practical algorithms is presented, applicable to monocular or stereo views. Effort has been made to present a balanced treatment of the theoretical and practical issues. Theoretical issues of broad and practical significance have been explored in detail, while other more specific ones are not covered. The topics covered in detail include image matching and motion estimation, error analysis, error estimation, stability, optimality, and experiments with simulated and real image data, with the objective being to present general techniques that have a mathematical foundation and engineering applications. This book is meant for graduate students and practicing engineers who are interested in computer vision. The organization of the material is intended to facilitate extraction and use of the algorithms without necessitating study of the rest of the material. Chapter 1 introduces the problem of motion analysis and briefly reviews past work. Chapter 2 is concerned with image matching, a step which is prerequisite for all further discussions in the book. Chapter 3 presents a closed-form solution for motion and structure parameters from two views of a rigid object. In the presence of noise, this solution is required as an initial condition for the optimization approach investigated in Chapter 4. The principles covered in Chapter 4 are used in later chapters whenever optimization is discussed. Chapter 5 and 6 address two separate issues, planar surfaces and line features for motion analysis, each of which can be skipped without loss of continuity. While Chapter 3 to 6 all assume monocular views, Chapter 7 is devoted to the analysis of image pairs acquired using stereo cameras. The techniques of robust statistics introduced in Chapter 7 can also be applied to monocular images. The subject of analyzing long image sequences, instead of only two or three views, is also treated in Chapter 7. Modeling and prediction of motion are covered in Chapter 8.