基于降质图像-光流场复原的智能轮椅测速

提出一种运动图像去模糊复原和基于仿射运动模型的光流场去抖动方法,以提高智能轮椅中光流里程计测速方法的精度。当轮椅线速度或角速度较大时,导致机载相机成像产生显著的运动模糊;且轮椅机器人的机械抖动也易产生光流场的偏差,进而影响速度估计的精度。针对于此,首先利用一种基于自适应模糊核的运动图像去模糊方法实现图像复原,以改善视频帧质量;其次,针对智能轮椅在行进中的机械抖动,利用随机抽样一致(RANSAC)排异后的光流场,在卡尔曼滤波框架下估计同名像点的仿射运动模型参数,进而实现光流补偿。实验结果表明所提方法能够提升基于光流场的智能轮椅视觉测速精度。

Velocity measurement of intelligent wheelchair based on restoration of degradated image and optical flow field

A method combining motion image deblurring restoration and mechanical de vibrating in optical flow field based on affine motion model is proposed to improve the accuracy of optical flow odometer based velocity measurement method for intelligent wheelchair. Large wheelchair translational velocity or rotational velocity leads to significant image motion blur for a fast moving onboard camera; additionally, the mechanical dithering of intelligent wheelchair robot easily deteriorates the quality of optical flow field; both of which affect the accuracy of velocity estimation. Aiming at this problem, a motion deblurring method based on adaptive fuzzy kernel is employed in this paper for image restoration and improving the quality of the video frames; Secondly, aiming at the mechanical vibration in the moving process of the intelligent wheelchair, under the framework of Kalman filter, the affine motion model parameters of consecutive image pairs are estimated with the optical flow field vectors refined by RANSAC (Random Sample Consensus), which realizes optical flow compensation for removing the mechanical vibration. Experiment results show that the proposed method is capable of improving the accuracy of visual velocity measurement of intelligent wheelchair based on optical flow field.