An Embedded Vision Engine (EVE) for automotive vision processing

This paper introduces Embedded Vision Engine (EVE) - a fully programmable, specialized vector processor architecture aimed at solving challenging Computer Vision applications encountered in Advanced Driver Assistance Systems (ADAS). The paper outlines the complexity of automotive vision applications, establishes why specialized architecture (like EVE) is needed and outlines the EVE architecture, its components and programming model. We present comparative benchmarks and provide an overview of many carefully crafted features of EVE for power management, inter processor communication, functional safety and software debug that helps in building a scalable, area-power efficient System-on-Chip (SoC) solutions for the cost, power and safety sensitive automotive vision space.