Optimization of computer vision algorithms in codesign methodologies
Egileak: Marcos Nieto, Juan Diego Ortega, Oihana Otaegui, Andoni Cortés
Data: 07.09.2014
Abstract
Embedding computer vision SW for in-vehicle applications requires the optimization of algorithms to fit into low cost and low consumption HW. Such optimization is a task substantially centered in improving the efficiency of the implementations, typically focused on the migration of algorithms to massive parallelization HW. The development cost associated with these actions can be prohibitive depending on the nature of the algorithms, which may be highly non-linear processes, iterative and recursive, partially sequential and often not fully parallelizable. As an alternative, co-design methodologies has shown to be a successful and feasible approach, which consists on coordinated work of the SW and HW development teams, with iterative evaluations of prototype SW implementations in reconfigurable HW platforms that allow detecting memory or processing bottlenecks. The analysis then guide the next iteration which may also consists on redesigning the algorithm and not only on implementation issues. In this paper we describe this general process applied to computer vision methods optimization. As a case study we present the improvements we have achieved designing a single camera vehicle detection system using the mentioned methodology to migrate it into an embedded platform (Xilinx Zynq 7000).
BIB_text
author = {Marcos Nieto, Juan Diego Ortega, Oihana Otaegui, Andoni Cortés},
title = {Optimization of computer vision algorithms in codesign methodologies},
pages = {1-12},
keywds = {
ADAS, computer vision, embedded systems, co-design, vehicle detection
}
abstract = {
Embedding computer vision SW for in-vehicle applications requires the optimization of algorithms to fit into low cost and low consumption HW. Such optimization is a task substantially centered in improving the efficiency of the implementations, typically focused on the migration of algorithms to massive parallelization HW. The development cost associated with these actions can be prohibitive depending on the nature of the algorithms, which may be highly non-linear processes, iterative and recursive, partially sequential and often not fully parallelizable. As an alternative, co-design methodologies has shown to be a successful and feasible approach, which consists on coordinated work of the SW and HW development teams, with iterative evaluations of prototype SW implementations in reconfigurable HW platforms that allow detecting memory or processing bottlenecks. The analysis then guide the next iteration which may also consists on redesigning the algorithm and not only on implementation issues. In this paper we describe this general process applied to computer vision methods optimization. As a case study we present the improvements we have achieved designing a single camera vehicle detection system using the mentioned methodology to migrate it into an embedded platform (Xilinx Zynq 7000).
}
date = {2014-09-07},
year = {2014},
}
