In the future, surgical visualization systems used in the field of microsurgical and endoscopic applications will be transformed into fully digital visualization. The surgeon observes the surgical field no longer via an optical lens system or video. The view is rather generated by virtual cameras, which is directed towards a highly accurate digital 3D reconstruction of the site. This 3D reconstruction can be regarded as the digital twin of the real surgical field. This virtual scene, which is updated for example, 60 times per second, is called Digital Reality (DR). The advantages of DR in surgical visualization are primarily its compatibility with other imaging modalities (CT, MRI, US, etc.), which also produce 3D objects. According to specific needs and applications, the information from these modalities can be fused with DR, enabling 3D-augmentation. Furthermore, the number of co-viewers and the location to use a DR visualization system is not restricted. Ultimately, DR forms the basis for the purely virtual operation simulations in the sense of optimized and personalized operation planning.
A key requirement for realizing DR is the instant and precise acquisition of spatially dense depth information needed for the real-time 3D reconstruction of the surgical situs. Due to the specific characteristics of surgical sceneries, such as limited color contrast and the difficult identification of distinct anatomical landmarks, conventional methods based on stereoscopic disparity may only lead to sparse and little reliable depth information.
To conquer this problem, we try to develop new methods for depth estimation from monoscopic video scenes based on depth cues known from human depth perception. Depth cues are for instance intrinsic image attributes like motion, perspective, texture, shading but also extrinsic cues like prior knowledge or experience. Ideally, those depth cues are suitable for both augmenting monocular endoscopic sceneries and also for complementing or supplementing sparse stereoscopic disparity maps.