Research On Deep Image Restoration Technology Based On Kinect

The appearance of Kinect makes the way of obtaining the depth information more convenient and quickly. So far, the application of Kinect equipment in the field of three-dimensional reconstruction, computer vision and other fields are more and more widely, more researchers have devoted to these fields. But due to the Kinect’s cheap inexpensive and other external factors, there are some errors when obtaining the depth information. These errors include black holes, noise and depth value stochastic jump and so on, and these errors would make the depth information to be inaccuracy. When doing work about three-dimensional reconstruction, it would affect the 3D reconstruction studies. This paper mainly researches on depth image inpainting technology and does the following studies to better meet the needs of the depth image, and makes the repaired depth image better applied to 3D construction research.This paper have analyzed the depth image inpainting research significance, through looking for large amounts of data, to understand the current research status about depth image inpainting at home and abroad, and has researched some key technologies. Then introduces the basic information of the depth map and the Kinect device, including depth image definition, the depth information and Kinect devices structure, application field, measuring errors’reason. It would be helpful for the sequent depth image restoration.Point at the problem of black hole in the depth image of Kinect equipment, this paper uses different methods to repair the depth image. For the black hole caused by the refraction and reflection of infrared, this paper propose an improved median filling algorithm based on the mean value and the median filtering method. For the black holes produced by occlusion, this paper proposes a filling method based on the area of the black hole size, if the area of a black hole is less than the set threshold value, using the neighborhood value to fill the black hole, otherwise using the improved bilateral filter which combined with color information to fill. Depending to the black hole caused by different reasons, and using the different methods to fill the black hole.At the same time, this paper analyzes the phenomenon of depth measuring error value at the situation when Kinect devices and objects are in two different states of acquiring depth information, combined with the idea of optimal estimation, this paper uses Kalman filter for depth images to resolve this issue, it can make the depth value jump smoothly. Finally, this paper propose the multi frame average method, using average method to confirm the final depth map, solve the problem of inaccurate depth value.Experimental result shows that the repair technology has achieved good results to get more accurate depth image information, it could be more conducive for the after work of 3D reconstruction.