针对飞行载体的实时ORB-SLAM实现问题,提出一种在嵌入式系统实现的改进ORB(oriented FAST and rotated BRIEF)单目视觉里程计算法.算法首先对输入图像进行灰度化、高斯滤波预处理实现简化运算和图像去噪,考虑到算法移植及在嵌入式系统实现,将图像预处理和ORB图像特征提取与匹配等功能封装为IP(intellectual property)核,布置到硬件系统中,提高特征提取与匹配的速度和正确率,保证位姿估计实时性.搭建ZYNQ嵌入式系统,开展对比实验,实验结果表明:改进后的算法特征点匹配率提高了3.78 倍,特征提取与匹配的耗时缩短为原来的1/8,处理图像的帧率可以达到19 fps,满足实时性要求.
Improved Implementation of ORB-SLAM in Embedded System
In response to the real-time ORB-SLAM implementation challenges on aerial platforms,an im-proved Embedded System implementation of the ORB(oriented FAST and rotated BRIEF)monocular vis-ual odometry algorithm is proposed.The algorithm first preprocesses raw images through grayscale conver-sion and Gaussian filtering to simplify computations and remove noise.With a focus on algorithm portabili-ty and embedded system implementation,functions such as image preprocessing,ORB feature extraction and matching are encapsulated into an intellectual property(IP)core and deployed on hardware systems to enhance feature extraction and matching speed and accuracy,ensuring real-time pose estimation.An embed-ded system based on ZYNQ is constructed,and comparative experiments are conducted.The experimental results demonstrate that the improved algorithm achieves a 3.78-fold increase in feature point matching rate,reduces the time consumption for feature extraction and matching to 1/8 of the original,and achieves a frame rate of 19 fps for image processing,meeting real-time requirements.
ORB-SLAMfeature extraction and matchingZYNQembedded systemFPGA
万方数据
维普
