基于ARM Cortex-A53 处理器平台,对 NEON 协处理器的指令和编程优化方法等技术进行了分析,并以此平台进行了D2C(Depth to Color)算法的优化实现。在树莓派3B+ SDK开发板上进行了充分实验,测试结果表明,NEON/VFP 协处理器在保持高的算法精度的同时可以较好地提高D2C算法的运算效率,并且保持低功耗、低成本。
Abstract
The NEON coprocessor instructions and programming optimization method based on ARM Cortex-A53 platform are introduced in this paper,and the optimization of D2C is also discussed.The D2C(Depth to Color) implementation is fully tested in Raspberry Pi 3B+ SDK board,and the results show NEON/VFP coprocessors improve the D2C calculation efficiency and keep good algorithm accuracy and low power and low cost.
关键词
ARM NEON /
SIMD /
D2C /
矢量化 /
树莓派3B+
Key words
ARM NEON /
SIMD /
D2C /
vectorization /
Raspberry Pi 3B+
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 胡新荣,王梦鸽,刘军平,等.基于Kinect的人体三维动作实时动态识别[J].科学技术与工程,2020,20(34) : 14133-14138.
[2] 张新钰,邹镇洪,李志伟,等.面向自动驾驶目标检测的深度多模态融合技术[J].智能系统学报,2020,15(4):758-771.
[3] 蒋罗.三维人脸重建与人脸识别[D].合肥:中国科学技术大学,2019.
[4] 蔡静怡,严飞,吴兆祥,等.基于Kinect v2的多视角三维重建实现[J].传感技术学报,2020,33(8):1149-1154.
[5] 刘帅.基于深度学习的3D视觉目标定位机械臂分拣系统[D].北京:北京林业大学,2019.
[6] 李坤.基于机器视觉的船体缺陷检测技术研究[D].大连:大连理工大学,2019.
[7] 计算机视觉:相机成像原理:世界坐标系、相机坐标系、图像坐标系、像素坐标系之间的转换[EB/OL].[2021-09].https://blog.csdn.net/chentravelling/article/details/53558096.
[8] 相机畸变与矫正[EB/OL].[2021-09].http://zhaoxuhui.top/blog/2018/04/17/CameraCalibration.html.
[9]深度图像配准(Registration)原理[EB/OL].[2021-09].https://www.cnblogs.com/cv-pr/p/5769617.html.
[10] ARM Limited.Cortex-A53: The Most Widely Used Low-Power Processor,2015.
[11] ARM Limited.ARM architecture reference manual ARMv8 for ARMv8a architecture profile beta,2016.
[12] ARM Limited.ARM NEON programmer’s Guide,2013.
[13] ARM Limited.ARM NEON intrinsics reference,2016.
PDF(1514 KB)
