3D Point Assignment Using 2D Bounding Polygons and Binning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 3D object recognition methods based on machine learning are computationally intensive and require significant manual effort for data annotation, making them inefficient and costly, especially for large data sets.
Innovation Solution
A method for assigning data points from three-dimensional data to two-dimensional bounding polygons using a projection matrix, followed by binning and clustering, which minimizes computational intensity and ensures efficient data processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning algorithms are used for 3D object recognition, then recognition accuracy is improved, but computational intensity and processing time increase significantly
Solution Approach 1:
The patent segments the 3D point cloud data into multiple two-dimensional projections (front, back, left, right views). Each projection is processed independently through the neural network, and results are combined. This segmentation reduces the computational complexity of processing entire 3D clouds while maintaining recognition accuracy through multi-view information fusion.
Solution Approach 2:
The patent transforms three-dimensional point cloud data into two-dimensional projection images. This dimensionality reduction converts complex 3D spatial data into 2D image formats that are more efficient for neural network processing, while still preserving sufficient geometric information for accurate 3D object recognition through the combination of multiple views.
2Measurement precision
If manual annotation is used for training data preparation, then data quality is improved, but time consumption and costs increase
Solution Approach 1:
The patent implements self-service annotation by using the neural network to automatically generate bounding boxes and labels for training data. The system processes raw sensor data through the trained model to produce annotated datasets without human intervention, enabling automated re-annotation when objects move or conditions change, thereby eliminating manual annotation time while maintaining consistent quality standards.
3Measurement precision
If large data sets are processed for training, then model accuracy is improved, but computational resources and processing time increase
Solution Approach 1:
The patent segments large 3D point cloud datasets into multiple smaller 2D projection views. This segmentation allows the neural network to process smaller, more manageable image datasets while still capturing comprehensive 3D object information through the combination of multiple perspectives, reducing memory requirements and processing complexity.
Solution Approach 2:
The patent converts large-volume 3D point cloud data into 2D projection images, significantly reducing the data dimensionality and storage requirements. This transformation enables efficient processing of large datasets through standard 2D image processing pipelines while preserving essential geometric features needed for accurate 3D object recognition.
Data Source
AI summary
A computer-implemented method for assigning data points from first data to one of several objects, wherein the first data comprises at least three spatial dimensions. Groups of projected data points are formed and assigned to the respective objects, taking into account received information data from two-dimensional bounding polygons and from a bin of the data point by an assignment method such that for overlapping polygons in which projected data points are located within more than one two-dimensional bounding polygon, the groups are formed such that no bin is assigned to more than one of the overlapping polygons at the same time. A device is also provided for data processing to execute the method, and a computer program product, and a computer-readable medium on which the above computer program product is stored.