Skip to content

Latest commit

 

History

History
executable file
·
101 lines (73 loc) · 3.98 KB

README.md

File metadata and controls

executable file
·
101 lines (73 loc) · 3.98 KB
Raw

3D Point Cloud Segmentation, Detection & Classification by PointNet in a Low-Cost System

alt text

System

image

Scripts Datasets
1.retrieveMiData.py MiData
2.retrieveMiDataCluster.py MiDataCluster
3.miDataClusterSegmentation.py ModelNet10
4.pointNet.py
5.segmentationClassification.py

Requirements ‼️

  • Python 3.6
  • Open3D
  • libroyale
  • Tensorflow
  • Keras
  • Trimesh
  • Scikit-learn
  • Seaborn
  • NumPy
  • Matplotlib

Usage ⚙️

Acquisition 📸

To capture 3D scenes using pmd Camboard pico flexx (MiData dataset):

python 1.retrieveMiData.py

To capture 3D scenes using pmd Camboard pico flexx filtered by confidence value (MiDataCluster dataset):

python 2.retrieveMiDataCluster.py

Segmentation ✂️

To segment, filter and cluster the scene:

python 3.miDataClusterSegmentation.py

Train 🧠

To train and evaluate a PointNet model:

python 4.pointNet.py

Classification 🗂

To segment and classify objects in a scene:

python 5.segmentationClassification.py

Results 📊

Acquisition

Depth Image
image

Point cloud
image

Segmentation

Different type of objects segmented
image image image

Train

Confusion matrix
image
Classification report
image
Multiple Predictions
image
Single prediction
image

Classification

Original point cloud
image
Clusters in filtered original cloud
image
Clusters in filtered point cloud
image
Cluster obtained after segmentation
image
Sampled cluster
image
Prediction
image

Bibliography 📖

Original PointNet implementation: https://github.com/charlesq34/pointnet
Original Keras implementation: https://github.com/keras-team/keras-io/blob/master/examples/vision/pointnet.py