Video conferencing usage dramatically increased during the pandemic and is expected to remain high in hybrid work. One of the key aspects of video experience is background blur or background replacement, which relies on good quality portrait segmentation in each frame. Software and hardware manufacturers have worked together to utilize depth sensor to improve the process. Existing solutions have incorporated depth map into post processing to generate a more natural blurring effect. In this paper, we propose to collect background features with the help of depth map to improve the segmentation result from the RGB image. Our results show significant improvements over methods using RGB based networks and runs faster than model-based background feature collection models.
In this paper, we devise a method that reduces distance errors in time-of-flight (ToF) images. Errors are exhibited at boundaries and surfaces that are not capable of reflecting the infrared ray. For the proposed method, at least two ToF cameras are required in the camera setup. ToF distance error region is estimated by comparing the captured ToF image with warped ToF image from the neighboring ToF camera. The distance values in the error region are replaced. A number of methods are examined to select the optimum replacement value. After distance error reduction, this method is inserted into the aforementioned depth map generation framework. The performance is analyzed by evaluating a synthetic image which is generated by the depth map result.
There are still have many serious problems with the real-existing scenes acquisition and generation of Hologram. In this research, an efficient CGH scheme that using orthographic projection images and depth map for real-existing scenes is proposed. The orthographic projection images and depth map are generated from 3D scanned model which is captured using depth camera. The proposed method generates Multiview images with full scanned real object with not only color information but depth information for hologram generation. The additional depth information can be used in additional artifact. This method reduces the number of angular samplings of the viewpoint images, provides all the human depth cues without producing any visible artifact.