/Multi-scale-pointcloud-registration

Pointcloud Registration Algorithm with noise and different scale management

Primary LanguagePythonMIT LicenseMIT

Registration Library OR-PCD 🚀

☁ OR-PCD is a library that implements a two block optimization approach in order to align Pointclouds. This method was created as a solution capable to manage scale and differences among the three coordinate axes of the clouds. In our approach we optimize the scale differences considering all the x/y/z coordinates directions as variables.

It integrates an inner optimization block where standard rototranslation matrix is obtained and an outer one where the best scale factors are estimated through with a pattern search algorithm ⠻⠏.

Library Structure

The library is composed of 3 main components and the objective is the let them be as extendible as possible in order to add other optimization pipelines during time. These are the components:

  • ⛓ Preprocessor: the preprocessor represent the standardization pipeline used before aligning the pointclouds. At the moment is composed of:
    1. Downsamplers: Porcess blocks that downsample pointclouds having too much points
    2. Scalers: Process block that scale the clouds initially
    3. Outliers: Porcess blocks that let you remove outliers from the cloud
  • 🔝 Optimizer: Inner block optimization methods. At the moment FastGlobal/GeneralizedICP are available
  • ⁑ The Aligner: is the outer block of the optimization. At the moment not extendible as it performs a pattern search with a multi-start procedure in order to find best scale parameters.

We provide some sample data to test and extend this library. You can find all of them within the data folder of the library.

Example code for beginners

First import main components

from or_pcd.Preprocessor.Downsamplers import RandomDownsampler
from or_pcd.Aligner import Aligner
from or_pcd.Optimizer import GeneralizedICP, FastGlobalOptimizer
from or_pcd.Preprocessor.Outliers import SOR
from or_pcd.Preprocessor.preprocessor import Preprocessor
from or_pcd.data import load_sample_cloud

Then load some sample clouds

source_cloud = load_sample_cloud("ArmadilloBack_330")
target_cloud = load_sample_cloud("ArmadilloBack_0")

Initialize optimizers with default parameters and run alignment

optimizer = FastGlobalOptimizer()
preprocessor = Preprocessor([RandomDownsampler(5000), SOR()])
aligner = Aligner(preprocessor, preprocessor, optimizer, visualize_intermediate_steps=True)
aligner.align(source_cloud, target_cloud)

Equally involved contributors:

Feel free to reach out to the contributors if you have any questions or suggestions.

Thank you for your interest in our project!