基于优化估计的深度图像修复与误差补偿方法研究

针对Kinect传感器在获取深度图像时存在深度值随机跳变的不准确性问题，基于最优估计的思想，提出卡尔曼滤波与多帧平均法相结合的图像修复方法。首先利用卡尔曼滤波对多幅深度图像进行修复处理，实现Kinect传感器在采集信息过程中随着时间递推，深度值的跳变逐渐趋于平稳的效果；然后基于多幅图像平均法确定最终的深度图像，解决了Kinect获取深度值存在误差导致的不精确问题。实验结果表明，该算法的均方根误差为38.102 5，平均梯度为0.471 3，信息熵为6.191 8，与单幅图像修复效果相比，得到的深度图像边缘更加清晰。

Repair and error compensation method for depth image based on optimization estimation

The depth value of Kinect sensor changes randomly when the depth image is obtained.In order to solve this problem, this paper presents an image repairing method combined with Kalman filtering and multiple frames averaging based on the idea of optimal estimation.Firstly, Kalman filter is used for repairing multiple depth images.The depth value tends to be stable with time recursion in the process of information capture by Kinect sensor.Secondly, multiple frames averaging method is used to determine the final depth image, in order to solve the problem of inaccurate depth value due to the error of Kinect sensor.The experimental results show that, the root mean square error of the algorithm is 38.102 5, the average gradient is 0.471 3, the information entropy is 6.191 8, the edge of depth image of this algorithm is more clearly when compared with the single image restoration.