Research On Kinect Depth Data Inpainting Algorithm

Depth camera data processing has been playing an important role in the field of graphics and augmented reality for a long time.With the appear of the cheap depth camera in recent years, important changes have taken place in depth camera field. Such devices which can acquire depth data with low price have the ability to collect data in real time.People of a lot of research and application field such as 3d reconstruction,tracking,recognition and segmentation are using these new type of equipment and puts forward a new strategy to solve many kinds of problems.But the data from these Kinect-like equipment have obvious noise and data defects.The disadvantages of these kind of data limits their application.A lot of People aimed at the phenomenon and put forward many different solutions.In this paper, wei analyzes Kinect’s imaging principle and discuss the characteristic and causes of the defect and inaccurate data. We conducted experiments and observation and then understand the problem of the Kinect’s data. We also introduces the commonly used strategies and basic theoretical knowledge of depth data in-painting. We compare these methods and tell the advantages and disadvantages of them.We put forward an in-painting method for Kinect specially. We use the Markov random field optimization framework to ensure the global optimal effect.We use the strategy of image pyramid and tensor voting to improve the traditional canny edge detection to apply to Kinect’s data.In the process of hole filling, we use meanshift segmentation algorithm joint with plane fitting.Finally we get some results which make us feel satisfied.