Scene Learning and Validation using Partial Observability

Summary

Mobile robotics has made great advances, however, still mobile robots have limited capabilities in context of understanding, recognizing and validating their environments. As an example, most mobile robots still represent the environment as a map with information about obstacles and free space.Some others enhanced this representation with information about relevant visual landmarks, but the semantic content is still missing. In order to enhance cognitive capabilities of mobile robots in indoor/outdoor environments, e. g. an office or a kitchen, we must provide them with a higher semantic understanding of their surrounding.

From a particular observation position only a limited part of a given 3D object can be sensed. For example, while being in front of an object back of the object is not visible to the sensor. A further complexity in recognition in 3D scenarios is that the system can approach a scene from an arbitrary direction. Hence the objects of the environment could be occluded when sensed from different view perspective. In order to solve the problem, objects have to be learned using their partial views that can be used later for recognizing those objects.

The objective of this research is to develop a mobile cognitive system able to learn and recognize different categories of objects in a scene using partial views which will be later used for scene validation. Achieving this ability in machine vision can lead to greater autonomy and intelligence in mobile cognitive systems.

Contact

• Prof. Dr.-Ing. Bärbel Mertsching