Container ROI Detection for Collision-Free Material Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing material handling systems face challenges in differentiating between products and container walls, leading to potential collisions and damage during fully autonomous operations.
Innovation Solution
A method involving 3-D vision data to classify regions-of-interest within containers, using 3-D point cloud data processing and machine learning to identify and navigate around container walls, preventing collisions by defining navigation paths.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 3-D point cloud data processing and machine learning classification are used to identify regions-of-interest, then measurement precision of container dimensions is improved, but device complexity increases
Solution Approach 1:
The container interior is segmented into multiple cluster areas based on 3-D point cloud data, with each cluster representing a distinct region. This segmentation enables precise measurement of container dimensions by analyzing the spatial distribution and boundaries of individual clusters, while breaking down the complex identification task into manageable regional units
Solution Approach 2:
The system transforms 3-D point cloud data into 2-D depth maps to simplify the analysis of container dimensions. This dimensional reduction maintains measurement precision by preserving depth information while reducing computational complexity, allowing the machine learning unit to efficiently classify regions without processing the full 3-D data structure
2Reliability
If navigation paths are defined to prevent collision with regions-of-interest, then reliability of material handling operations is improved, but device complexity increases
Solution Approach 1:
The system performs preliminary classification of regions-of-interest using machine learning before material handling operations begin. By pre-identifying container walls and structural elements through 3-D point cloud analysis and 2-D depth map classification, the navigation path can be pre-planned to avoid these regions, ensuring reliable operations without requiring complex real-time collision detection and response mechanisms
Solution Approach 2:
The 2-D depth map serves as an intermediary representation between the 3-D point cloud data and the navigation path planning. This intermediate 2-D format simplifies the communication of spatial information to the path definition unit, reducing the complexity of navigation control while maintaining the reliability needed to prevent collisions with identified regions-of-interest
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The disclosed embodiments relate to a material handling method that generates three-dimensional (3-D) point cloud data based on a field-of-view of an image capturing unit. A first set of cluster areas are extracted from a plurality of cluster areas based on orientation data of the 3-D point cloud data. Further, a two-dimensional depth map is generated based upon the 3-D point cloud data. A candidate region that corresponds to a cluster area from the first set of cluster areas is determined. A ratio of a cross-sectional area of the cluster area and a cross-sectional area of the container is determined that exceeds a first cross-sectional threshold. Accordingly, a classification score of the candidate region is determined when the determined ratio exceeds a first cross-sectional threshold. In response to classifying the candidate region as a region-of-interest, a navigation path in the container that prevents collision with the region-of-interest is defined.