Neural Network Voxel Feature Transformation for Point Cloud Robustness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous driving systems face challenges in accurately recognizing objects and surroundings due to the variability and quality of point cloud data from LiDAR, which affects the robustness of deep learning models.
Innovation Solution
A processor-implemented method that determines either a first voxel or a first key point of point cloud data and performs feature transformation through a neural network to determine a second voxel or a second key point, where the features of the second voxel and key point differ from the first, and these are used for training the point cloud data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If large and diverse point cloud data is used for training, then model robustness is improved, but data quality and variability are insufficient
Solution Approach 1:
The patent uses a neural network to generate synthetic point cloud data that copies and transforms real LiDAR data. The network learns from actual point cloud data and generates augmented training data with varied features, effectively creating high-quality training samples without requiring extensive physical data collection.
Solution Approach 2:
The neural network transforms point cloud data by changing parameters such as voxel features and key point features. The system modifies data characteristics through feature transformation, creating diverse training samples with varied parameters while maintaining the underlying structural information.
2Adaptability or versatility
If feature transformation is performed through neural network, then data variability is improved, but training data quality may be compromised
Solution Approach 1:
The patent employs a feedback mechanism where the neural network's generated data is evaluated and used to refine subsequent generation. The system continuously improves the quality of transformed features by learning from the results of previous transformations, ensuring that data variability increases while maintaining quality standards.
Solution Approach 2:
The neural network performs preliminary feature extraction and transformation before the main training process. By pre-processing the point cloud data to identify and transform key features in advance, the system ensures that the transformed data maintains high quality while achieving the desired variability for robust training.
Data Source
AI summary
A processor-implemented method includes determining either one or both of a first voxel and a first key point of point cloud data, and by performing feature transformation on either one or both of the first voxel and the first key point through a neural network, determining either one or both of a second voxel and a second key point of the point cloud data, wherein a voxel feature of the first voxel is different from a voxel feature of the second voxel, a key point feature of the first key point is different from a key point feature of the second key point, and either one or both of the second voxel and the second key point are used for training of the point cloud data.