Research On RGB-D Data Alignment Algorithm Based On Microsoft Kinect

In recent years, the depth image has been widely used in human-computer interaction, 3D reconstruction. The general depth sensor,such as Microsoft Kinect, can only achieve limited precision.The resultant depth image are often found to be noisy,misaligned with the color image,and even contain many large holes. These limitations make it difficult to be adopted by many depth graphics applications. In this paper,we increases the quality of the limited depth image by using image processing.To address the defects of the original depth image,such as noise pollution and large holes.We provide different solutions to different types of defects. In this paper,we change the weight of joint bilateral filtering to solve the noise pollution of depth image. Experimental results show that the algorithm has better effect for image noise problem. Then,we redefine the weight coefficient of the diffusion pixels in the fast marching method to fill the large holes. The result of experiment and optimization results validate the feasibility of the proposed optimization algorithm.The paper also gaves a global optimization algorithm of the depth image based on the Markov random field.The new optimization framework is given by expanding the constraints of the traditional Markov Random Field.The optimization framework can fuse features of both color and depth images,and can produce high-quality consistent depth.This paper proposes an effective method to evaluate the confidence of depth image,and uses it for adaptive weighting of MRF data constraints; then,we use the modified fast marching method to get the hole constraints; thirdly,we use the edge information and the segmentation result to get the edge-sensitive smoothness constraints.At last, the experiment results show the feasibility of this algorithm.To evaluate this algorithm, we compared this algorithm with Park’s method,and the result show this algorithm achieves much higher precision than Park’s method.