Tests were run with the initial input depth equals 1 which was successively increased by 1.
By inputting depth prior information or 3D models, the methods in [6, 7] can also restore haze-free images.
To obtain a more consistent rendered sequence, a temporal filtering procedure for the input depth sequence is beneficial.
The Structural SIMilarity index (SSIM) [37] is utilized to distinguish the stationary scene from the moving foreground objects by combining the input depth images.
By considering the spatial relationship provided by the input depth maps, the texture information of these occluded stationary scenes can also be obtained.
The proposed method incrementally merges segments obtained from each input depth image in an unified global model leveraging the camera pose estimated via SLAM.
In the control network, the reinforcement learning procedure, the trained model mentioned above was used to forward every input depth image of real time and get the feature maps of the depth image.
These applications require even better accuracy of depth imaging than in the case of 3D TV, since the successful operation of various classification or motion analysis algorithms depends on the quality of input depth features.
In detail, the pre-processing stage selectively filters the input depth map to reduce the bit-rate cost while minimizing quality degradation, and the post-processing restores the reconstructed depth map as much as possible.
The 3D structure of the environment is reconstructed by combining obtained depth maps in the voxel space, and camera motion is estimated by the ICP algorithm using an estimated 3D structure and the input depth map, which is depth-based vSLAM.
To evaluate this claim, we have designed a pattern classifier for three different inputs: (1) depth map, (2) texture map, and (3) both depth and texture maps of the facial surface.
