<title>Three-dimensional registration using range and intensity information</title>

A review of recent range image registration methods with accuracy evaluation

“…Although in some situations all points have the same weight [24], in other situations weights are introduced depending on: distance between point-correspondences [29], compatibility of normals [29] and the uncertainty of the covariance matrix [30], among others.…”

A Review of Point Cloud Registration Algorithms for Mobile Robotics

“…While some papers present general algorithms assuming that a reasonable transformation will be given [Armesto et al, 2010, Druon et al, 2006, Schutz et al, 1998, Jost and Hügli, 2002, others include humans in the loop [Godin et al, 1994, Pulli, 1999. More integrated solutions rely on external sensor (wheel encoders, Inertial Measurement Unit (IMU), Global Positioning System (GPS), etc.)…”

“…The motivation for the selection of those registration techniques is to reduce local minima possibilities with coarse alignment methods and continue with more precise but computationally more expensive methods. From the papers reviewed, the system can have two [Bosse and Zlot, 2008, Censi, 2008, Godin et al, 1994, Stewart et al, 2003, Tsai et al, 2010 or three layers [Pan et al, 2010, Bosse andZlot, 2009b], but nothing seems to limit the number of layers implemented on a given system.…”

A robust and fast method for 6DoF motion estimation from generalized 3D data

“…Once the matches have been established, outliers should be detected in order to improve the results. We use a similar approach like the ones used in Godin et al (1994); Pulli (1999); Rusinkiewicz and Levoy (2001) in which each match is weighed by the inverse of the distance between the points involved in the match, but in the present paper, both object position and orientation are used to compute this distance. Other approaches used to reject invalid matches based on some compatibility test Turk and Levoy (1994); Masuda et al (1996); Dorai et al (1998) also depend on the 3D sensor used.…”

“…However, the amount of information discarded by these techniques is quite important and may affect the accuracy of the results. The addition of some constraints, such as color similarity Godin et al (1994); Pulli (1997) or surface normal vectors Pulli (1999), in the matching step can improve the consistency of the matched pairs and thus the final alignment computed by the ICP. The main problem in these methods is that these constraints depend on the kind of sensor used.…”