PROBABILITY LINEAR METHOD POINT CLOUD APPROXIMATION

PROBABILITY LINEAR METHOD POINT CLOUD APPROXIMATION

Paper in proceedings (conference paper)

Fitting curves through point clouds is useful when the further computation is required to be fast or the data set is too large. The most common method to fit a curve into a point cloud is the approximation using the Least squares method (LSM) but it can be used only when the expected data have normal distribution. Data obtained from LIDAR often tend to have an error which can’t be solved by LSM, like data shifted in one angular direction. The main goal of this paper is to propose more efficient method for estimation of obstacle position and orientation. This method uses curve approximation based on probability; this can solve some classic errors that appear when processing data obtained by LIDAR. This method was tested and was found to have a disadvantage: great demand for computing power; its more than ten times slower than classic LSM and in cases with normal distribution gives the same results. It can be used in system where the emphasis is on accuracy or in multiagent solution when working with big data set is not desired.

Fitting curves through point clouds is useful when the further computation is required to be fast or the data set is too large. The most common method to fit a curve into a point cloud is the approximation using the Least squares method (LSM) but it can be used only when the expected data have normal distribution. Data obtained from LIDAR often tend to have an error which can’t be solved by LSM, like data shifted in one angular direction. The main goal of this paper is to propose more efficient method for estimation of obstacle position and orientation. This method uses curve approximation based on probability; this can solve some classic errors that appear when processing data obtained by LIDAR. This method was tested and was found to have a disadvantage: great demand for computing power; its more than ten times slower than classic LSM and in cases with normal distribution gives the same results. It can be used in system where the emphasis is on accuracy or in multiagent solution when working with big data set is not desired.

Point cloud, LIDAR, Curve approximation, Laser range finder, Localization

Point cloud, LIDAR, Curve approximation, Laser range finder, Localization

KRÁLÍK, J.; VENGLÁŘ, V.

2021

24.11.2020

978-80-214-5896-3

ENGINEERING MECHANICS 2020 26th INTERNATIONAL CONFERENCE

1st edition, 2020

306

309

4

https://www.engmech.cz/im/im/download/EM2020_proceedings.pdf