The work presented here is about employing a theory of updates to study geometrically observable changes that occur in spatial information about image sequences of a dynamic scene. The logical framework consists of a formalism for specifying the geometrical content of a scene, as well as the changes that occur in this geometry, and an algorithm for constructing a description for such changes from logical deductions. In this approach, a database state represents the available sensor data at a particular time instant. Transitions in sensor data are modeled by changes in the database and interpreted based on axioms encoding commonsense spatial reasoning. The main contribution of this work is that it provides the theoretical foundations for symbolically interpreting long sequences of sensor data transitions. For testing the framework and its implementation, the problem of interpreting rotational movements of objects in a sequence of images was used. Our experiments show that the system correctly interprets rotational movements for objects of different colors and provides satisfactory results for interpreting such movements from perceptually indistinguishable objects.
- Knowledge representation
- Machine vision
- Qualitative spatial reasoning
- Reasoning about actions and change