The purpose of the D2C algorithm is to align depth image and color image both generated by depth camera at the same moment,its real-time implementation on the embedded platform is an important step of 3D application miniaturization.The CEVA XM4 DSP based on AR9201 platform is characterized by low power consumption and high computing capability.Combined with the instruction features of CEVA DSP,a series of methods are proposed for D2C algorithm optimization to meet the requirements of important engineering indicators.The engineering practice results show that the optimized D2C algorithm performs well in the system engineering environment,D2C frame rate of 960P resolution image reaches 30 fps on average after algorithm DSP vectorization.The proposed optimization methods also provided a reference resolution for the real-time implementation of other image processing algorithms on similar embedded platform.
Key words
CEVA DSP /
SIMD /
D2C /
vectorization /
multi-core parallel processing
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
References
[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):14133-14137.
[2] 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.
[3] 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.
[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):149-1154.
[5] 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].[2021-12].https://blog.csdn.net/chentravelling/article/details/53558096.
[6] 相机畸变与矫正[EB/OL].[2021-12].https://zhaoxuhui.top/blog/2018/04/17/CameraCalibration.html.
[7] 深度图像配准原理[EB/OL].[2021-12].https://www.cnblogs.com/cv-pr/p/5769617.html.
[8] CEVA.Ceva-xm4_arch_spec_vol-i,2017.
[9] CEVA.Ceva-xm4_arch_spec_vol-ii,2017.
[10] Artosyn.Ar9201_product_brief.v1.3,2018.
PDF(1695 KB)
