next up previous
Next: Thesis

Discovering visual features of the environment and building 3D map from a single camera motion

Dipak Mahadeo Suman Jogle (Y7111010)

Supervisor: Dr. Amitabha Mukerjee

Abstract

Simultaneous localization and mapping (SLAM) has traditionally relied on laser range finders and the use of vision from SLAM is relatively new. Approaches such as MonoSLAM (Single camera SLAM) have suffered from two problems: (a) manual camera calibration at start, (b) tracking loss after relatively short period. In this work, we developed CSLAM (Calibration integrated SLAM) which address problem (a) by providing as auto-initialisation mechanism based on the homography computed for a known planar 2D object.

For the problem of track-loss, we use two-view geometric and calculate the motion of the camera between two consecutive frames. Whenever the calibration object is visible, we have adequate number of features for our calculation at any frame in the scene. The present approach is limited to only the situations where the calibration object was continuously visible. The approach provides a mechanism for direct camera calibration resulting in improved camera trajectory.



next up previous
Next: Thesis
Dipak Jogle 2009-09-22