Existing automotive safety vision systems are deployed
with basic functionality such as lane detection, pedestrian
detection and vehicle detection using data fused with other
sensors. Some systems are standalone vision systems with
very minimal vision tasks such as video overlay for rear-view
parking guidance. The major bottleneck in integrating all the
above applications into a single platform is providing realtime
results. Also, with such integrated platforms, it is not
guaranteed that we achieve 100% efficiency with accurate
performance under all types of road and weather conditions.
This again calls for dedicated algorithms and therefore high
throughput to handle such unique conditions. A specific need
arises for two things; 1) the need for hardware that performs
similar algorithmic tasks with increased computing power to
evaluate the algorithm performance of the existing hardware
and 2) the evaluation of hardware with a parallel computing
architecture as a prototype to handle vision algorithms [1] in a
automotive environment. A large set of algorithms integrated
into one vehicle-mounted, high-performance computing
platform with can lead to problems with V-SLAM, motion
planning and sensor fusion, features that are most pivotal in
autonomous vehicle research.
Existing automotive safety vision systems are deployed
with basic functionality such as lane detection, pedestrian
detection and vehicle detection using data fused with other
sensors. Some systems are standalone vision systems with
very minimal vision tasks such as video overlay for rear-view
parking guidance. The major bottleneck in integrating all the
above applications into a single platform is providing realtime
results. Also, with such integrated platforms, it is not
guaranteed that we achieve 100% efficiency with accurate
performance under all types of road and weather conditions.
This again calls for dedicated algorithms and therefore high
throughput to handle such unique conditions. A specific need
arises for two things; 1) the need for hardware that performs
similar algorithmic tasks with increased computing power to
evaluate the algorithm performance of the existing hardware
and 2) the evaluation of hardware with a parallel computing
architecture as a prototype to handle vision algorithms [1] in a
automotive environment. A large set of algorithms integrated
into one vehicle-mounted, high-performance computing
platform with can lead to problems with V-SLAM, motion
planning and sensor fusion, features that are most pivotal in
autonomous vehicle research.
การแปล กรุณารอสักครู่..
