The byline for this ehapter eloquently summarises our approach to motion estimation: it must happen quickly and it must happen without regard to any ancillary description of the environment. To avoid a moving object, all that is required is to know its location and approximate velocity knowing what the object is. its colour or its features is not necessary in this process. We treat motion as a fundamental quantity that is measured as directly as possible. This naturally leads to the gradient based methods, where the only intermediate description of the visible world consists of the intensity derivatives. Alterna tive motion estimation techniques require more complex intermediate descriptors: frequency domain techniques require sets of filter banks or Fourier transforms, and token based methods require explicit extraction of some type of structure from the world. This chapter draws together gradient based motion estimation, the rigid body motion model, dynamic scale space and a set of environmental assumptions to create a simple motion estimation algorithm. The resulting algorithm determines quickly the piecewise projection of relative three-dimensional translational motion onto the camera's X axis.

Inspec keywords: gradient methods; channel bank filters; motion estimation; frequency-domain analysis; navigation; Fourier transforms

Other keywords: three-dimensional translational motion; autonomous navigation; gradient based motion estimation; filter banks; Fourier transforms; token based methods; frequency domain techniques

Subjects: Radionavigation and direction finding