Analysis and Calibration of a Mirror Setup producing Mirror-reflected, Multi-view Videos

Passive stereo photogrammetry determines 3D surface images only based on the images taken by two or more cameras without the projection of a pattern. Due to the missing, projected pattern, the process of finding correspondences between views/images is more difficult and ambiguous, usually surface details are used in this case.
Hybrid stereo photogrammetry combines both active and passive methods in order to achieve higher accuracy and quality in 3D surface imaging.

There are many 3D surface-imaging technologies that are often used in multiple medical applications, some of those technologies are described below, for a detailed analysis please refer to Tzou et al., 2014.

3dMD is a dynamic 4D system (3D plus time) introduced in 2004. This technology exploits hybrid stereo photogrammetry using both projected random patterns and texture of the skin (pores, freckles, etc.) to stereo triangulate and generate a 3D surface image. The software algorithms is used both for 3D reconstruction and calibration. System calibration takes up to several minutes depending on the hardware set-up. The system is made by two or three high frame rate machine vision cameras, with a PC-controller workstation.

Axisthree is focused on 3D simulation using clinical data to do physics-based tissue-behavior simulation on models. In 2006, Axisthree created a technology called color-coded triangulation together with Siemens where the colored pattern simplifies the finding of correspondences. Axisthree uses the principal of structured light to create 3D surface images. System calibration takes less than 5 min, and it is necessary only when the hardware has been moved. The system is made by three or four high frame rate machine vision cameras, with a PC-controller workstation and a projector.

Canfield exploits the principle of passive stereo photogrammetry, where the texture of the skin is used to determine the geometry and generate a 3D surface image. Canfield supplies four hardware options. Depending on the chosen hardware setup, the system can be used to capture: 100◦ of frontal faces-face, neck and décolletage-face, breast and body. There is also a portable, customized and versatile version of this system for 3D acquisition. The provided software is used to perform tissue simulations, or can be used to simulate multiple surgical and non-surgical facial procedures, also provides automatic breast measurements and simulates breast augmentation outcomes. The system uses dedicated hardware for the purpose.

Crisalix is the first web based 3D simulator for plastic surgery and aesthetic procedures. Unlike the other companies, Crisalix does not offer any hardware. The program creates 3D surface images from three 2D images taken with a consumer camera, physical distance measurements of the patient’s anatomy and a set of landmarks. Crisalix offers two products. 3D MAMMO simulator is used for planning and biomechanical simulations of breast implants and skin elasticity while 3D FACE simulator is used for surgical and non-surgical facial procedures. The user needs a pc/laptop and a standard consumer camera (digital camera, webcam, smartphone, etc.)

Di3D is a 4D system, which captures 3D video sequences of dynamically changing surfaces. Di3D’s technology exploits passive stereo photogrammetry. System calibration takes about 5 min. They offer custom capture systems with up to 32 cameras, while the basic system is made by four cameras.

 
The above described systems are cost intensive, and are usually made by a projector and at least two cameras, so those system usually need a lot of space and request some synchronization requirements. The synchronization between the cameras is needed because of the tracking operation. This means that you cannot have any of the previously described systems at home. In this thesis we refer to a simpler system described by Frey et al., 1999 which proposes to use a system of two mirrors to simulate stereoscopic vision, this system will be described in detail in the later paragraphs.