The objective of this research is image sequence analysis of complex objects. The term complex is here used for describing that the object simultaneously articulates and deforms. The goal is to provide descriptors that are related to the problem of understanding human behaviour from image sequences, which can be used for identification with many possible law enforcement and defence applications. This, however, involves several challenging problem areas of which a few have been addressed in this thesis. Although the focus has been on recovering the shape and motion of humans, another object-class has been incorporated. Tanks are chosen as they appear with both intravariations and moveable parts.

Many potential applications motivated the initialisation of this work. One is the lack of analytical methods for the identification of a crime suspect on surveillance video recordings. A comprehensive survey of methods and work for the criminal justice system is one part of this thesis. A case study is included in this part to demonstrate the usefulness of 3D modelling and motion estimation. Another driving force has been to extend the area of automatic target recognition and identification in airborne surveillance applications to include deformable and articulated objects. An identification system incorporates several procedures. Imaging technologies are one important part. The new development in active camera systems is an appealing area and a work on a system for gated viewing for target identification at long range is included. To identify objects from an image sequence, they first have to be detected and tracked. It is a well-known fact that video generates a rich set of information and the motivation for the work on detection and tracking of moving objects has been to reduce the search space and extract useful data for further processing, such as identification. Finally, a framework and a method to simultaneously estimate the 3D shape and motion pattern of complex objects, such as the human body, are proposed and discussed. The work expands and combines algorithms originally developed in the area of model-based coding of faces to recover object deformations, with algorithms for estimating object articulations.