Minimally invasive surgery is a quite challenging task to the surgeon due to operation through an endoscope and sensitive telescopic instruments exclusively. Computer assisted stereo endoscopic systems eases depth perception and supports several tasks such as dissection of a renal tumour by augmented reality. An essential procedure is robust surface reconstruction of the observed organs. Due to strong reflections from the endoscopic light source, homogeneous textures and weak deforming geometries robust surface reconstruction becomes quite challenging and is not solved successfully yet. Therefore, in this work a novel miniaturised three camera endoscope is introduced and an algorithm for three image analysis and semi-global optimisation is implemented. Synthetic and real experimental measurements are conducted to evaluate theoretical assumptions and review three hypotheses. In contrast to stereo analysis, it is examined whether three image analysis leads to more robust results, allows for smaller matching window sizes and replaces a time-consuming semiglobal matching algorithm. The investigations show that the first two assumptions can generally be confirmed, but the semi-global matching is necessary in some cases. Additionally, errors by reflections are examined in more detail and are suppressed efficiently by crossed polarising filters. The novel three camera endoscope and customized image analysis algorithm gives a great benefit to computer assisted endoscopy and brings research a step closer to more reliable assistant systems.