Research On Depth Video Enhancement Algorithm Based On Tensor

With the continuous improvement of human cognitive ability and demand,traditional images,grayscale video,color video and so on have been unable to convey all the information that people's need.The appearance of depth camera is an innovation in computer vision,image processing and other fields.Depth data collected by depth sensor can contain depth information of the object,which is not affected by light and surface unequal.The theory of low rank matrix recovery is a new topic in the field of machine vision and artificial intelligence in recent years.Low rank matrix recovery derives from the theory of compressed sensing.After years of research,it has gradually become an independent system and extends to the low rank tensor recovery,which can handle Gao Weishu's basis.Due to the problem of the device itself or the influence of external conditions,the depth data collected by the depth camera usually contain noise and voids.The aim of this paper is to denoise and repair the depth video captured by Kinect.The existing video denoising methods are mostly based on frames,and the internal structure of video can not be maintained in the form of matrix.In order to solve these problems,this paper proposes a tensor based depth video enhancement algorithm.First,we preprocess the depth video obtained by Kinect,use the weighted moving average mechanism to suppress the flickering effect,and separate the foreground and background from the processed depth video,then only enhance the foreground moving target,and most of the noise can be eliminated during the separation process.Then we match the similar block with the depth video only with the foreground object,and use the tensor to express each similar block,keep the internal structure of the video information,find the corresponding color video blocks in the aligned color video frames for the similar blocks in each depth video,and use the joint bilateral filtering to guide the color image and video blocks.The edge of the image.After weighted average of the processed similar blocks,the original video is put back,and then the tensor recovery model is used to repair and remove the whole video noise,so as to achieve the purpose of holistic hole filling.A large number of experiments show that the proposed method has a good effect on the denoising and restoration of depth video,and maintains the internal structure of video.