Optimising computer vision based ADAS: vehicle detection case study
Abstract
Computer vision methods for advanced driver assistance systems (ADAS) must be developed considering the strong requirements imposed by the industry, including real-time performance in low cost and low consumption hardware (HW), and rapid time to market. These two apparently contradictory requirements create the necessity of adopting careful development methodologies. In this study the authors review existing approaches and describe the methodology to optimise computer vision applications without incurring in costly code optimisation or migration into special HW. This approach is exemplified on the improvements achieved on the successive re-designs of vehicle detection algorithms for monocular systems. In the experiments the authors observed a ×15 speed up between the first and fourth prototypes, progressively optimised using the proposed methodology from the very first naive approach to a fine-tuned algorithm.
BIB_text
title = {Optimising computer vision based ADAS: vehicle detection case study},
journal = {IET Intelligent Transport Systems},
pages = {157-164},
number = {3},
volume = {10},
keywds = {
Object detection; driver information systems; computer vision; low consumption hardware; code optimisation; computer vision based ADAS optimization; monocular systems; advanced driver assistance systems; low cost hardware; vehicle detection algorithms<
}
abstract = {
Computer vision methods for advanced driver assistance systems (ADAS) must be developed considering the strong requirements imposed by the industry, including real-time performance in low cost and low consumption hardware (HW), and rapid time to market. These two apparently contradictory requirements create the necessity of adopting careful development methodologies. In this study the authors review existing approaches and describe the methodology to optimise computer vision applications without incurring in costly code optimisation or migration into special HW. This approach is exemplified on the improvements achieved on the successive re-designs of vehicle detection algorithms for monocular systems. In the experiments the authors observed a ×15 speed up between the first and fourth prototypes, progressively optimised using the proposed methodology from the very first naive approach to a fine-tuned algorithm.
}
isi = {1},
doi = {10.1049/iet-its.2014.0303},
date = {2016-03-28},
year = {2016},
}
