Direkt zum Inhalt | Direkt zur Navigation

UniBwM » LRT » Technik Autonomer Systeme » Technik Autonomer Systeme (TAS) » Forschung » Geschichte (english!) » Expectation-based, Multi-focal, Saccadic (EMS) Vision system

Expectation-based, Multi-focal, Saccadic (EMS) Vision system

Expectation-based, Multi-focal, Saccadic (EMS) Vision system

Based on the experience gained up to 1996, a new (third) generation machine vision system was to be developed, combining the advantages of active visual / inertial and multi-focal sensing with the 4-D approach exploiting spatio-temporal models. Dynamic vision is regarded as an animation process driven by phases of visual observation and an internal complementary production of motion- and action- processes. These latter ones are based on knowledge about typical motion processes of objects and about behaviors and decision making of subjects; subjects are defined as objects with the additional capability of perceiving (part of) the environment and of generating actions on their own.

The own system (body plus sensors, interpretation processes, typical control outputs and actuators) is the subject represented internally in most detail; only for the own body are inertial measurement signals and direct knowledge about control outputs for the actuators available. The control outputs are the results of the execution of standard behaviors, which are necessary for performing maneuvers (e.g. lane changing or turning-off onto a crossroad for ground vehicles). At any time the ego subject is performing one of these maneuvers, which are selected by decision processes in order to realize a mission. A first description of the EMS-vision system to some detail may be found in [IV 2000] with the contributions [Gr et al 00, GrD 00, Ho et al 00, Lü D 00,Mau 00, PeD 00]. At the example of the "Turning-Off" -maneuver, representation, selection and execution of a maneuver in the EMS-Vision system are described in [SiD 00].

For other subjects, most of these quantities have to be derived from observation of body motion by vision; rather crude stereotypical behaviors for performing maneuvers predominate in interpretation of actions of other subjects observed.

In the field of application "land vehicle guidance", the EMS-Vision system has been implemented on the two test vehicles of ISF, VaMoRs and VaMP. For both vehicles the hardware architecture (EMS-Vision hardware, below) is the same.

EMS-Vision hardware for land vehicles:


Figure 10: EMS-Vision hardware block diagram

The essential set of components of the EMS-Vision hardware is a network of commercial off-the-shelf (COTS-) dual processor PCs in an industrial casing. The computing power of the system can be increased very simply by adding new PCs to the network or by replacing PCs by more efficient ones. All real time data are communicated via a SCI-Net (Scalable Coherent Interface). Via a second net (FastEthernet) the "Gateway PC" is connected to all other PCs. In this way all PCs can access the system software that is placed on the "Gateway PC" for a consistent software management. All "Image-Processing PCs", grabbing video streams from different cameras (CCD-, Low Light Level- or High Dynamic Range-cameras), are connected with one terminal for visualization. The "Gateway PC" and the "Behavior PC" are connected with another terminal for system operation. All other subsystems of the vehicles for real-world interfaces to sensors and actuators are linked by special interfaces to the "Behavior PC". These are either transputer systems remaining from the previous generation or more modern special processors.