Real Time Panoramic Depth Imaging Using Standard Cameras

摘要

A computer vision is a special kind of scientific challenge as we are alludusers of our own vision systems.udOur vision is definitely a source of the major part of information weudacquire and process each second.udA stereo vision is perhaps even greater challenge, since our own visionudsystem is a stereo one andudit performs a complex task, which supplies us with 3D information on ourudsurroundings in a very effectiveudway.ududMaking machines see is a difficult problem. On one side we haveudpsychologicaludaspects of human visual perception, which try to explain how the visualudinformation is processedudin the human brain. On the other side we have technical solutions, which tryudto imitateudhuman vision. Normally, it all starts with capturinguddigital images that store the basic informationudabout the scene in a similar way that humans see. But this informationudrepresents only theudbeginning of a difficult process. By itself it does not reveal theudinformation about the objects on the scene,udtheir color, distances etc.udto the machine. For humans, visual recognition is an easy task, but theudhuman brain processing methods areudstill a mistery to us.ududOne part of the human visual perception is estimating the distances to theudobjects on the scene.udThis informationudis also needed by robots if we want them to be completely autonomous.ududIn this dissertation we present a stereo panoramic depth imaging system.ududThe basicudsystem is mosaic-based, which means that we use a single standard rotatingudcamera and assemble the captured images in a multiperspective panoramicudimage.udDue to a setoff of theudcamera's optical center from the rotational center of the system we areudable to capture the motion parallax effect, which enables the stereoudreconstruction. The camera is rotating on a circular path with the stepuddefined by an angle equivalent to one-pixel column of the captured image.udTo find the corresponding points on a stereo pair ofudpanoramic images the epipolar geometry needs to be determined.udIt can be shown that the epipolar geometry is very simple if we are doingudthe reconstruction based on a symmetric pair of stereo panoramic images.udWe get a symmetric pair of stereo panoramic images when we takeudsymmetric columns on the left and on the right side from the capturedudimage center column.ududThis system howeverudcannot generate panoramicudstereo pair in real time. That is whyudwe have suggested a real time extension of the system,udbased onudsimultaneously using manyudstandard cameras.udWe have not physically built the real time sensor, but we have performedudsimulations to establish the quality of results.ududBoth systems have beenudexhaustively analysed and compared. The analyses revealed a number ofudinterestingudproperties of the systems.udAccording to the basic system accuracy we definitely can use the system forudautonomous robot localisation and navigation tasks.udThe assumptionsudmade in the real time extension of the basic systemudhave been proved to be correct,udbut the accuracy of the new sensor generally deteriorates in comparison toudthe basic sensor.ududGenerally speaking, theuddissertation canudserve as a guide for panoramic depth imagingudsensor design and related issues.udud