A 3D real-time vision system based on passive stereovision algorithms: Application to laparoscopic surgical manipulations

Although 3D-vision systems can be an interesting way for precise laparoscopic surgical manipulations, the few existing ones present the disadvantages of giving only qualitative rather than quantitative 3D information. In this paper we present a first step of the stereo laparoscope which gives a real time 3D scene dimensions. It is based on algorithms coming from mobile robotics, called "passive stereo-vision" and that can achieve the 3D-vision compatible with the surgical acts execution times. Due to the evolution of a particular kind of electronic circuits called FPGA (Field Programmable Gate Array) which include a lot of multipliers, macro-cells and memories, it is possible today to develop a passive stereovision based architecture with real time performance and giving an output rate up to one hundred images per second. The algorithm which we have adopted to be implemented on the FPGA, in order to reach such performances, is the Correlation Census. This one has been chosen for its simplicity and its parallelism aspect. The main objective of this algorithm is to search the correspondence between two input images (right and left) taken from two different angles to get a "depth map" from which it is possible to reconstruct the 3D scene. To compute this algorithm, many pipelined stages of pre-processing are needed (means calculus, windowed Census transformation, best correspondence searching, and imagefiltering...).