# Classification of Point Cloud Data in Road Scenes Based on PointNet++

## TL;DR

This paper improves point cloud classification in road scenes using PointNet++ with effective data augmentation and preprocessing techniques.

## Contribution

A novel point-filling preprocessing method and optimized PointNet++ MSG framework for high-accuracy road scene classification.

## Key findings

- Farthest Point Sampling and rigid transformations improve dataset quality and model performance.
- Point filling outperforms zero-padding with 86.49% training and 98.23% test accuracy.
- The optimized model achieved 97.41% test accuracy using PointNet++ MSG with tailored hyperparameters.

## Abstract

What are the main findings?
Effective Dataset & Augmentation Strategies: Farthest Point Sampling preserves features better than random sampling, rigid transformations enhance diversity, and noise injection improves authenticity. Point filling significantly outperforms zero-padding (86.49% train/98.23% max test acc vs. 66.86%/79.89%).Optimized Model Performance: Using optimal hyperparameters (lr = 0.00075, batch = 6, Adam, PointNet++ MSG), the model achieved 86.26% avg train acc (98.54% max) and 97.41% test acc. Most categories (e.g., biker, excavator) performed excellently, while a few (e.g., building, traffic lights) had low recall due to sample issues; minor misclassifications from small dataset/imbalance were mitigated by hyperparameter tuning.

Effective Dataset & Augmentation Strategies: Farthest Point Sampling preserves features better than random sampling, rigid transformations enhance diversity, and noise injection improves authenticity. Point filling significantly outperforms zero-padding (86.49% train/98.23% max test acc vs. 66.86%/79.89%).

Optimized Model Performance: Using optimal hyperparameters (lr = 0.00075, batch = 6, Adam, PointNet++ MSG), the model achieved 86.26% avg train acc (98.54% max) and 97.41% test acc. Most categories (e.g., biker, excavator) performed excellently, while a few (e.g., building, traffic lights) had low recall due to sample issues; minor misclassifications from small dataset/imbalance were mitigated by hyperparameter tuning.

What are the implications of the main findings?
This study presents a viable high-precision classification technique for autonomous driving and map creation, providing a reliable reference for related research and applications in complex road environments.The findings provide offers a reusable framework for road scene point cloud dataset construction, data augmentation, and PointNet++ modification, supporting similar 3D data processing tasks.

This study presents a viable high-precision classification technique for autonomous driving and map creation, providing a reliable reference for related research and applications in complex road environments.

The findings provide offers a reusable framework for road scene point cloud dataset construction, data augmentation, and PointNet++ modification, supporting similar 3D data processing tasks.

Point cloud data, with its rich information and high-precision geometric details, holds significant value for urban road infrastructure surveying and management. To overcome the limitations of manual classification, this study employs deep learning techniques for automated point cloud feature extraction and classification, achieving high-precision object recognition in road scenes. By integrating the Princeton ModelNet40, ShapeNet, and Sydney Urban Objects datasets, we extracted 3D spatial coordinates from the Sydney Urban Objects Dataset and organized labeled point cloud files to build a comprehensive dataset reflecting real-world road scenarios. To address noise and occlusion-induced data gaps, three augmentation strategies were implemented: (1) Farthest Point Sampling (FPS): Preserves critical features while mitigating overfitting. (2) Random Z-axis rotation, translation, and scaling: Enhances model generalization. (3) Gaussian noise injection: Improves training sample realism. The PointNet++ framework was enhanced by integrating a point-filling method into the preprocessing module. Model training and prediction were conducted using its Multi-Scale Grouping (MSG) and Single-Scale Grouping (SSG) schemes. The model achieved an average training accuracy of 86.26% (peak single-instance accuracy: 98.54%; best category accuracy: 93.15%) and a test set accuracy of 97.41% (category accuracy: 84.50%). This study demonstrates successful road scene point cloud classification, providing valuable insights for point cloud data processing and related research.