3D Point Cloud Label Estimation Using Sum of Sigmoid Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for assigning labels to three-dimensional point clouds, such as those described in Patent Literature 1 and Non-Patent Literature 1, face limitations in accuracy due to the use of human-designed features and the lack of explicit utilization of shape features like point normals, especially when dealing with point clouds having a low number of representative points.
Innovation Solution
A three-dimensional point cloud label learning and estimation device that clusters point clouds using a neural network, employing geometric transformations and the total sum of sigmoid function values for feature extraction, which incorporates attribute information including point normals to improve label assignment accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning features are used for label assignment, then identification accuracy is improved, but the system cannot accurately identify shapes with a low number of representative points
Solution Approach 1:
The patent combines deep learning features with explicitly extracted shape features (normals and geometric attributes) to create a hybrid feature representation. This merging allows the system to leverage the power of deep learning while maintaining robustness for low-point clouds through explicit geometric feature extraction and sum-of-sigmoid processing that aggregates information across all available points.
Solution Approach 2:
The patent changes the feature extraction parameter from max pooling (which selects only the maximum value) to sum of sigmoid function values (which aggregates information from all points). This parameter change ensures that even when the number of representative points is low, the system can accumulate sufficient geometric information through the summation process, improving reliability for identifying shapes with fewer points.
2Device complexity
If max pooling processing is used in feature extraction, then processing simplicity is maintained, but shape features are not effectively utilized for identification
Solution Approach 1:
The patent changes the aggregation parameter from max pooling to sum of sigmoid function values in the feature extraction layer. This parameter change transforms the feature extraction mechanism to explicitly utilize shape features by summing transformed geometric attributes (normals, curvatures) across all points, thereby improving shape identification accuracy while maintaining a relatively simple processing framework.
3Productivity
If human-designed histogram features are used, then computational simplicity is achieved, but identification accuracy is limited
Solution Approach 1:
The patent substitutes the mechanical feature design process (human-designed histogram features) with an automated deep learning-based feature extraction system. The neural network automatically learns and extracts relevant features from the point cloud data, including geometric attributes like normals and curvatures, replacing manual feature engineering with automated learning while maintaining computational efficiency through vectorized operations.
Solution Approach 2:
The system enables self-service feature extraction where the deep learning model automatically identifies and extracts relevant features from the input point cloud without requiring manual feature design. The model adapts to different point cloud characteristics and automatically determines which geometric features (normals, curvatures, positions) are most relevant for accurate identification.
Data Source
AI summary
A three-dimensional point cloud label learning and estimation device includes: a clustering unit that clusters a three-dimensional point cloud into clusters; a learning unit that makes a neural network learn to estimate a label corresponding to an object to which points contained in each of the clusters belong; and an estimation unit that estimates a label for the cluster using the neural network learned at the learning unit. In the three-dimensional point cloud label learning and estimation device, the neural network uses a total sum of sigmoid function values (sum of sigmoid) when performing feature extraction on the cluster.