Point Cloud is a cloud of 3D points representing a particular object,Scene,etc.
This is a research Project which Focuses on classifying the respective points to its respective object cloud.There are 2 methods of doing this namely- 1.Segmentation classification 2.point by point classification
This project basically focuses on point by point classification.
The libraries needed for the processing are-
1.PyntCloud-https://github.com/daavoo/pyntcloud
2.Numpy
3.Pandas
4.Sklearn
5.Matplotlib.pyplot
6.Anaconda and Jupyuter Notebook
7.Open3D
Dataset Used-Paris Rue Madam Dataset(http://cmm.ensmp.fr/~serna/rueMadameDataset.html)
The algorithms used for classification in the basic model-Random Forests
Note:The Better library and Documentation for Point Cloud manipulation in Python is PCL-Python binding(https://github.com/strawlab/python-pcl).If C++ then the PCL library will be the best.
The PreProcessing Steps which can be done For the Project is-
1.Voxel Grid Downsampling or Filtering
The Feature Extraction From the point Cloud is-
1.Eigen Values
2.Anisotrophy
3.Curvature
4.EigenTrophy
5.EigenSum
6.Linearity
7.Omnivariance
8.Planarity
9.Spehericity
The further Feature extraction that can be included are-
1.Slope
2.Height horizontal and vertical
3.Angular Features
4.Moments Calculation
The post Processing techniques which can be used is-
1.Shape Distribution histogram
2.Graph cuts
This model gives an accuracy of 92% which can be improved by feature tuning and the above stated methods