Depth Inpainting Algorithm of RGB-D Camera Combined with Color Image
Yi Zhang, Jianchun Dai, Han Zhang, Lida Yang · 2018
I416 order to solve the problem of poor quality of deep image of RGB-D cameras, a depth inpainting algorithm based on combined color image information is proposed. First of all, In order to align color image with deep image, the RGB-D camera is calibrated by the Zhang's demarcation method. Then, it is marked as hollow and noise based on the size of the connected domain which depth value is zero in the depth image. Using the domain variance method to classify the hollows. The domain similarity and color consistency algorithm are proposed for two types of hollows to be repaired. Finally, the local bilateral filtering method is used to remove the noise and ensure the original depth value of the depth information as much as possible. The experimental results show that the repaired depth image can effectively repair the hollow, which is used for the target 3D positioning, and the positioning accuracy is higher.