2009/01/01 by Olaf Kähler, Kähler, Olaf, Joachim Denzler +1
Computer Science · Engineering · #3d reconstruction #Advanced Vision and Imaging #Optical measurement and interference techniques #Robotics and Sensor-Based Localization #Structure-from-motion #tracking
paper · doi:10.4230/dagsemproc.08422.3
openalex publication_date 2009/01/01 · openalex created_date 2017/07/31 · openalex updated_date 2026/07/28
The structure-from-motion problem is central in applications like visual robot navigation and visual 3d modeling. Typical solutions split the problem into feature tracking and geometric reconstruction steps. Instead we present a combined solution, where the tracking step is implicitly supported by a feedback of 3d information, and the geometric reconstruction is statistically optimal in case of Gaussian noise on image intensities. Experiments confirm an increased accuracy and reliability, and despite a significant computational overhead, the combined solution still runs at 5-10 fps.