Visual SLAM: sparse, dense and inertial aided mapping
A consistent estimation of the 3D structure of the environment with areliable localization (SLAM problem) is a fundamental requirement forproviding a mobile robot with autonomous navigation capabilities.Using vision to achieve this task (Visual SLAM) offers the benefit ofproviding plenty of information (geometric and photometric) relevant toanalyze the perceived scene.In this talk, we present methods to tackle the Visual SLAM problem inboth the sparse and the dense map representation cases.In the sparse setting, we propose an epipolar geometry-based VisualOdometry framework designed to address the specific problem of motionestimation robust to motion blur, and a stochastic filtering approach tothe visual-inertial navigation problem that moreover allows automaticcalibration between the camera and a low-cost IMU. The IMU intrinsiccalibration problem is also faced, proposing an easy to implement procedurethat well approximates the classical, high cost, mechanical procedure.In the dense setting, we present a system able to infer the densestructure of the environment, represented by a piecewise smooth surface,along with the ego-motion of the camera by performing a robust trackingof the projection of this surface in the image. Short BioAlberto Pretto is one of the members of the Intelligent AutonomousSystems Laboratory (IAS-Lab) of the University of Padua since 2003.Pretto's main research interests are in the field of Robot Vision andRobot Navigation. He received his "Laurea" in Computer Science at theUniversity of Padua (Italy) in 2003 and in 2009 he got his Ph.D.degree in Information Engineering from the University of Padua. In2004 and 2005, he has been working as software engineer at PadovaRicerca Scpa. In 2005, he is one of the funders of IT+Robotics Srl, aspin-off company of the University of Padua working on robotics andmachine vision. From 2009 to 2010, and from 2011 he is a post-doc research fellow at the Department of Information Engineering of the University of Padua. He is currently doing research on robot visual exploration, visual SLAM and object recognition, in cooperation with the UCLA VisionLab of prof. Stefano Soatto, Los Angeles (USA), where he spent a 9 months visiting research fellowships between 2011 and 2012.