嵌入式板上实时运行的数据流ORB-SLAM算法
需积分: 14 184 浏览量
更新于2024-09-04
收藏 912KB PDF 举报
"数据流ORB-SLAM在嵌入式GPU板上的实时性能优化"
数据流ORB-SLAM是为了在嵌入式GPU板上实现实时的SLAM（Simultaneous Localization and Mapping）算法，满足机器人、无人驾驶飞机和地面车辆等嵌入式板上的应用需求。由于嵌入式板的硬件限制，直接移植台式机CPU设计的算法是不现实的，因此需要对SLAM算法进行修改和自定义，以适应嵌入式板的计算资源限制。
本文中，我们介绍了如何修改和自定义开源SLAM算法ORB-SLAM2，以便在NVIDIA Jetson TX2上实时运行。我们采取数据流范式来处理图像，获取有效的CPU/GPU负载分配，从而获取每秒约30帧的处理速度。在KITTI数据集的四个不同序列上的定量实验结果证明了该方法的有效性。
数据流ORB-SLAM算法的关键点在于：
1. 数据流范式：我们使用数据流范式来处理图像，实现高效的CPU/GPU负载分配。这种方法可以实时地处理图像，获取高帧率的处理速度。
2. ORB-SLAM2算法修改：我们对开源SLAM算法ORB-SLAM2进行了修改和自定义，以适应嵌入式板的计算资源限制。我们优化了算法的计算流程，减少了计算复杂度，提高了算法的实时性。
3. 嵌入式板上的实现：我们在NVIDIA Jetson TX2上实现了数据流ORB-SLAM算法，获取了每秒约30帧的处理速度，满足了机器人、无人驾驶飞机和地面车辆等嵌入式板上的应用需求。
数据流ORB-SLAM算法的优点在于：
1. 高实时性：数据流ORB-SLAM算法可以实时地处理图像，获取高帧率的处理速度，满足了机器人、无人驾驶飞机和地面车辆等嵌入式板上的应用需求。
2. 高效性：我们使用数据流范式来处理图像，实现高效的CPU/GPU负载分配，减少了计算复杂度，提高了算法的实时性。
3. 可扩展性：数据流ORB-SLAM算法可以扩展到不同的嵌入式板上，满足了不同应用场景的需求。
数据流ORB-SLAM算法是一种高效、实时的SLAM算法，满足了机器人、无人驾驶飞机和地面车辆等嵌入式板上的应用需求。
Data Flow ORB-SLAM for Real-time Performance
on Embedded GPU Boards
Stefano Aldegheri
1
, Nicola Bombieri
1
, Domenico D. Bloisi
2
, and Alessandro Farinelli
1
Abstract— The use of embedded boards on robots, including
unmanned aerial and ground vehicles, is increasing thanks to
the availability of GPU equipped low-cost embedded boards in
the market. Porting algorithms originally designed for desktop
CPUs on those boards is not straightforward due to hardware
limitations. In this paper, we present how we modified and
customized the open source SLAM algorithm ORB-SLAM2 to
run in real-time on the NVIDIA Jetson TX2. We adopted a
data flow paradigm to process the images, obtaining an efficient
CPU/GPU load distribution that results in a processing speed of
about 30 frames per second. Quantitative experimental results
on four different sequences of the KITTI datasets demonstrate
the effectiveness of the proposed approach. The source code of
our data flow ORB-SLAM2 algorithm is publicly available on
GitHub.
I. INTRODUCTION
Navigation is the main task for an autonomous mobile
robot. In order to move from the current position A to a
desired destination B, a mobile robot needs a map, to know
its position on the map, and to have a plan to get from A to
B, possibly selecting the most appropriate from a number of
alternative routes. Simultaneous Localization and Mapping
(SLAM) aims at processing data coming from robot sensors
to build a map of the unknown operational environment and,
at the same time, to localize the sensors in the map (also
getting the trajectories of the moving sensors).
Many different types of sensors can be integrated in
SLAM algorithms such as laser range sensors, encoders,
inertial units, GPS, and cameras. In recent years, SLAM
using cameras only has been actively discussed because
cameras are relatively cheap with respect to other sensor
types and their configuration requires the smallest sensor
setup [1]. When the input for SLAM is visual information
only, the technique is specifically referred to as visual SLAM
(vSLAM).
vSLAM algorithms can be grouped according to three
categories, namely feature-based, direct, and RGB-D ap-
proaches. In feature-based methods, geometric information
from images is estimated by extracting a set of feature
observations from the image in input and by computing
the camera position and scene geometry as a function of
these feature observations only [2]. Direct (or featureless)
approaches aims at optimizing the geometry directly on the
image intensities and use photometric consistency over the
1
Stefano Aldegheri, Nicola Bombieri, and Alessandro Farinelli are with
the Department of Computer Science, University of Verona, Strada le Grazie
15 - 37134 Verona, Italy nicola.bombieri@univr.it
2
Domenico D. Bloisi is with the Department of Mathematics, Computer
Science, and Economics, University of Basilicata, Viale dell’Ateneo Lucano,
10 - 85100 Potenza, Italy domenico.bloisi@unibas.it
Fig. 1. (a) NVIDIA Jetson TX2 module. (b) The KITTI dataset [5]
whole image as an error measurement. In RGB-D SLAM,
dense depth enables the detection of planes that have no
textures [3].
One of the main challenges in vSLAM is to achieve real-
time processing. Direct and RGB-D methods are computa-
tionally demanding and requires GPU computation to run
in real-time. ORB-SLAM2 [4] is, at the moment, the most
complete feature-based vSLAM system [1]. It works in real-
time on standard CPUs, but not on embedded boards.
In this paper, we present a modified version of ORB-
SLAM2 that runs in real-time on an NVIDIA Jetson TX2
embedded board (see Fig. 1). According to [6], ORB-
SLAM2 is the package that provides the best result in
terms of accuracy among the most popular sparse methods.
However, it is a high-demanding algorithm in terms of CPU
and memory usage [7], thus a careful computational load
distribution is required to obtain real-time performance with
hardware limitations.
The contributions of this work are three-fold:
1) We use a data flow paradigm to obtain a representation
of the original algorithm as a graph, which allows to
subdivide efficiently the computational load between
CPU and GPU.
2) Experimental results demonstrate that, by balancing
CPU/GPU usage, it is possible to achieve real-time
performance on four different sequences of the KITTI
dataset while maintaining good accuracy.
3) We provide on GitHub the complete source code
optimized for real-time use on the NVIDIA Jetson
TX2.
2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
Macau, China, November 4-8, 2019
978-1-7281-4003-2/19/$31.00 ©2019 IEEE 5370
下载后可阅读完整内容，剩余5页未读，立即下载
2021-10-10 上传
2021-03-08 上传
2021-05-14 上传
2021-09-29 上传
2021-03-31 上传
2021-03-23 上传
2021-02-04 上传
2021-05-19 上传
izh20
- 粉丝: 18
- 资源: 1
我的内容管理
展开
最新资源
- protGear:protGear是在进行主要分析之前用于蛋白质微阵列数据处理的软件包
- Excel模板多媒体课件统计表.zip
- 第二周作业：第二周作业
- twitter:（）–用于在Twitter上自动:cyclone:更新媒体和:artist_palette:艺术作品的插件
- Excel模板大学优秀学生申请校内专业调整拟录取名单公示.zip
- statistical_rethinking
- HxgcIDReader_20180821.rar
- bookmanage
- CloudSimPerSimple
- Story:我的杰作
- Excel模板大学学期教学进程计划.zip
- gtk-js-app:标准GtkGNOME JS应用程序的模板
- 离子项目
- 2014-2020年扬州大学341农业知识综合三考研真题
- chat-app
- typescript-rest-api:该存储库需要
