Point Cloud Object Augmentation for Occlusion-Aware Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object-augmentation technologies randomly apply data augmentation techniques to all objects without considering their characteristics or geometry, leading to a deterioration in data quality, especially in high-occlusion scenarios.
Innovation Solution
An object detection apparatus that performs attribute-aware data augmentation by selecting source and target objects based on their characteristics and geometry, and applies data augmentation techniques such as swapping or mixing points between these objects to generate augmented point cloud data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If existing object-augmentation technology performs probabilistic random augmentation for all objects, then the amount of training data is increased, but data quality deteriorates especially in high-occlusion scenarios
Solution Approach 1:
The patent applies different augmentation strategies to different objects based on their local characteristics. Objects are classified into categories (e.g., foreground/background, occluded/non-occluded) and subjected to appropriate augmentation techniques. This ensures that data quality is preserved for critical objects while still increasing data quantity through selective augmentation.
Solution Approach 2:
The patent changes the parameters of data augmentation by considering object attributes such as occlusion level, position, and importance. Instead of uniform random augmentation, the system adjusts augmentation intensity and type based on object-specific parameters, thereby maintaining data quality while increasing training data quantity.
2Ease of operation
If data augmentation is applied without considering object characteristics and geometry, then processing simplicity is maintained, but data quality deteriorates
Solution Approach 1:
The patent segments objects into different categories based on their characteristics and geometry information. By dividing objects into groups (e.g., based on occlusion attributes, spatial relationships), the system can apply simplified processing rules to each segment while maintaining overall data quality. This segmentation approach balances processing simplicity with quality preservation.
Solution Approach 2:
The patent performs preliminary analysis of object characteristics and geometry before applying data augmentation. By pre-classifying objects and determining their augmentation requirements in advance, the system maintains processing simplicity during the actual augmentation phase while ensuring data quality through informed decision-making.
3Adaptability or versatility
If global-augmentation technology is used to rotate, reverse, or zoom scene data, then data diversity is increased, but the ability to process data for each object or class is lost
Solution Approach 1:
The patent combines global-augmentation techniques with object-level processing by segmenting the scene into individual objects. This allows the system to apply global transformations to the entire scene while also performing object-specific augmentations, thereby maintaining both data diversity and fine-grained processing capability.
Solution Approach 2:
The patent merges global-augmentation and object-augmentation approaches. By combining scene-level transformations with object-level operations, the system achieves both data diversity through global changes and processing precision through object-specific treatments, resolving the contradiction between versatility and complexity.
Data Source
AI summary
An object detection apparatus obtains point cloud data. The object detection apparatus selects a source object and a target object based on characteristics of multiple objects included in the point cloud data and geometry information between the multiple objects. The object detection apparatus selects a target partition to apply data augmentation based on geometry information between the source object and the target object. The object detection apparatus performs data augmentation of the point cloud data by applying an augmentation technique to the selected target partition, and outputs the augmented point cloud data.