Abstract:
Three-Dimensional (3D) scene reconstruction using depth cameras
is ubiquitous in Augmented Reality, Robotics, and Medical Imaging.
Although many gradient-based highly computational reconstruction methods have been proposed in the literature, there is hardly
any attempt at using meta-heuristic optimization techniques like
the Genetic Algorithm (GA) for performing global camera pose
estimation in a scene reconstruction framework. In this paper, we
develop a 3D scene reconstruction framework that uses a combination of local image features and outlier removal technique
(RANSAC) for performing local camera pose estimation. Further,
we formulate a GA based global camera pose estimation approach.
The 3D model is represented using an efficient and salable voxelbased representation. The paper presents the influence of various
parameters on the local and global camera pose estimation techniques and the prescription for best-suited parameter values to
achieve near-optimal performances
