Autonomous road vehicle guidance in normal traffic

Abstract

A passenger car Mercedes 500 SEL has been equipped with the sense of vision in the framework of the EUREKA-project ‘Prometheus III’. Road and object recognition is performed both in a look-ahead and in a look-back region; this allows an internal servo-maintained representation of the entire situation around the vehicle using the 4-D approach to dynamic machine vision. Obstacles are detected and tracked both in the forward and in the backward viewing range up to about 100 meters distance; depending on the computing power available for this purpose up to 4 or 5 objects may be tracked in parallel in each hemisphere. A fixation type viewing direction control with the capability of saccadic shifts of viewing direction for attention focussing has been developed. The overall system comprises about 45 transputers T-222 (16-bit, for edge extraction and communication) and T-805 (32-bit, for number crunching and knowledge processing) and 4 boards based on the Motorola Power Chip (MPC-601) for obstacle detection including image segmentation and state estimation. A description of the parallel processing architecture is given; system integration follows the well proven paradigm of orientation towards 4D physical objects and expectations with prediction error feedback. This allows frequent data driven bottom-up and model driven top-down integration steps for efficient and robust object tracking.