Developing Integrated Vision Applications for Active Safety Systems 2009-01-0158

Current image-processing solutions are limited with respect to being simultaneously flexible, scalable, high-performing and efficient. As vision-based safety systems increase in functionality and become more widely adopted, system makers will require flexible and scalable solutions to accommodate the needs of an expanding market. To support the highly complex embedded automotive vision systems of the future, therefore, engineers are now turning to dedicated vision processors in place of standard off-the-shelf solutions.

This paper will describe how to develop highly integrated image processing systems for active safety applications using the unique capabilities of a highly parallel, reconfigurable SIMD-MIMD processor architecture that offers the ability to handle both single- and multi-core designs. This architecture will enable safety systems to execute multiple applications simultaneously to provide more comprehensive driver assistance information. Several sensor sources will be discussed, including radar, vision, light detection and ranging (LIDAR), in addition to common advanced safety applications such as lane tracking and obstacle detection. The paper will also describe how the reconfigurable architecture allows for more robust and efficient algorithm development, as the unique performance of the SIMD-MIMD architecture supports real-time processing of images even when using highly complex algorithms.