Harmonization between Stereoscopic VR Images

Providing realistic immersive experiences through virtual reality (VR) has become a major goal of video technology. Techniques to generate monocular VR images have been extended to the stereoscopic mode by adding the sense of depth. Research on the generation of stereoscopic VR images relies on either specialized equipment or general-purpose cameras. We propose an efficient algorithm to generate stereoscopic 360 VR images from pictures captured using affordable conventional cameras. To overcome the limited ability of conventional cameras to harmonize left and right VR images, we propose a model that represents the relation between left and right views using Euler angles.

The following figure illustrates the generation of stereoscopic 360 VR images using two cameras in parallel to take pictures from left and right views. The left and right cameras capture the difference between the views, and they remain fixed by the rig binding them.

The following figure shows examples of left and right panoramic images generated by the conventional techniques. The red dashed lines indicate the different warping results for the left and right panoramic images due to the random sample consensus and independent optimization during bundle adjustment.