Deploying and Scheduling Vision Based Advanced Driver Assistance Systems (ADAS) on Heterogeneous Multicore Embedded Platform

Abstract

In-vehicle electronics is the fastest growing area of auto technology. A large part of this growth is in the area of Advanced Driver Assistance Systems (ADAS), which have become a very important part of the modern automobile. These ADAS systems usually run multiple applications simultaneously (e.g., Lane Departure Warning, Traffic Sign Detection, Pedestrian Detection, etc.). Each of these applications consists of multiple tasks and has hard real-time constrains that need to be satisfied. Finding new ways to reduce the cost, size, weight, complexity, and power consumption of the hardware used in these systems, while being able to offer more and more configurable features has always been a challenging problem for automotive suppliers and OEMs. The solution can be found in heterogeneous multicore embedded System on Chip (SoC), but the transition toward multi-core architectures has created more challenges at the software side, where fast deployment, integration and verification of the software is required. In addition to that, the temporal behavior of such safety critical applications must be predicted, with minimal processor reservation and memory usage. In this paper we will present a workflow to easily deploy and schedule multiple vision based ADAS applications on heterogeneous multi-core platform. This will be accomplished by employing Single Rate Data Flow (SRDF) graph as a base model of computation combined with hierarchical scheduling strategy that provides functional isolation and predictable timing between these applications. The results of this work could be used towards the design and implementation of partitioned, real-time operating systems that execute on heterogeneous multicore system on chip environment and a tool-chain to auto-generate such a system.