Research On Depth Image Enhancement Technology Based On Time-of-flight Three-dimensional Camera

Time-of-flight(TOF)depth camera is a kind of device with depth image acquisition,and has much attention in recent years.It is widely used in the field of computer vision.For the image obtained by TOF camera has the defects such as larger random noise and lower resolution,the corresponding image enhancement algorithm must be used to get a depth map with high-resolution and quality for subsequent use.The depth image enhancement method in this paper includes the following two parts: De-noising of depth image and super-resolution reconstruction of depth image.In the de-noising part,this paper selects a multi-image averaging method to reduce the noise influence of the depth image after analyzing and comparing several de-noising algorithms.Firstly,take the multiple sequence depth images of the same scene,then make the depth maps sum together.Finally,obtain the mean value and reduce the image detail fuzzy interference caused by random noise,increase the practicability of the depth mapIn the resolution enhancement part of the image,the low-resolution depth image is up-sampled in the following two aspects: the method of using the output images of the depth camera only,and the method of combining the color image in same scene.In the process of up-sampling with the output images of TOFdepth camera,firstly,the contents of some interpolation algorithms and their limitations are listed.Then a weighted interpolation algorithm based on confidence and edge optimization is proposed in this paper.Enhance the weight of the high-confidence depth information in the interpolation process,and improve the traditional interpolation algorithm.Finally,according to the image gradient multi-direction auto-correlation feature information,the edge pixel band of the image is optimized,so that the obtained high-resolution depth map is greatly improved from the visual effect and the performance index.In conjunction with the depth up-sampling of the same scene,the color texture information is used as the prior knowledge constraint to assist the depth image to complete the super-resolution reconstruction.For the defect with the distortion of depth value after reconstruction in the region of depth continuous relatively has gray-scale step.This paper proposes a second-order total generalized variation algorithm based on edge-optimized interpolation.The low-resolution depth map is pre-processed by an edge-optimized interpolation algorithm before the energy function model construct,and bring more restraint for the fidelity of data reconstruction.Experiments show that this method can improve the accuracy of the depth value obtained by up-sampling and has a clearer edge.