Photogrammetric computer vision
The Photogrammetric Compter Vison (PCV) Lab and its collaborators at UC Berkeley, MIT and NASA Glenn Research Center developed an astronaut navigation system for landed space missions. The system consisted of multiple integrated sensors, including a stereo camera, IMU and step sensor, to precisely locate an astronaut and navigate him/her to a target science experiment site.
Under the direction of Professor Alper Yilmaz and with the support of Center for Surveillance, the National Science Foundation, Air Force Research Labs and industry partners, PCVLab developed a photogrametry suite to recover 3D scene from an aerial multihead camera-mount. The camera mount is consisted of 6 cameras with minimal overlapping fields of view.
Given multiple images per epoch, the system generates a virtual camera view by generating a mosaic. The matching features across each mosaic are utilized to generate a sparse, three dimensional model by a bundle adjustment process. This step is followed by a dense 3D recovery which takes appearance similarities and sparse 3D into account.