Object Recognition Using Range Image Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object recognition techniques face challenges such as high dimensionality of data, reliance on texture, sensitivity to occlusion, instability of normal features, and limited recognition of non-flat objects, leading to inefficient processing and suboptimal results.
Innovation Solution
An object recognition method using a range image-based approach that detects interest points, extracts features, matches them with a preliminarily-created object model, calculates a transformation matrix, and verifies its validity to recognize objects robustly, even under occlusion and varying viewpoints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If high-dimensional features are extracted from 3D point clouds for object recognition, then recognition robustness is improved, but processing time increases and real-time performance deteriorates
Solution Approach 1:
The patent segments the high-dimensional 3D point cloud data into multiple two-dimensional range images taken from different viewpoints. Instead of processing all 3D data at once, the system divides it into manageable 2D slices, each processed independently for feature extraction and matching, thereby reducing overall processing time while maintaining recognition robustness through multi-view information.
Solution Approach 2:
The patent transforms 3D point cloud data into 2D range images, effectively reducing dimensionality from three dimensions to two. This dimensionality reduction enables faster processing using 2D image processing algorithms while preserving essential geometric information needed for robust object recognition through multiple viewpoint representations.
2Productivity
If contour-based recognition is used for fast calculation, then processing speed is improved, but recognition accuracy deteriorates under occlusion
Solution Approach 1:
The patent segments the object representation into multiple contour views from different viewpoints rather than relying on a single contour. Each contour is processed quickly using 2D processing, but the combination of multiple contours provides robustness against occlusion, as occlusion affecting one viewpoint does not prevent recognition from other viewpoints.
Solution Approach 2:
The patent uses multiple 2D range images from different viewpoints instead of a single 3D representation or single contour. This multi-view 2D approach maintains fast 2D processing speeds while improving occlusion robustness by providing alternative views where occluded regions may be visible from other angles.
3Reliability
If 3D point cloud data is processed directly, then recognition robustness is improved, but device complexity and computational requirements increase
Solution Approach 1:
The patent converts 3D point cloud data into 2D range images, reducing computational complexity by working in two dimensions rather than three. This transformation maintains essential geometric information for robust recognition while enabling the use of simpler, faster 2D processing algorithms and reducing memory requirements.
Solution Approach 2:
The patent creates multiple 2D copies (range images) of the 3D object from different virtual viewpoints rather than directly processing the full 3D point cloud. These 2D copies serve as simplified representations that retain sufficient information for recognition while being computationally easier to handle.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
To carry out satisfactory object recognition in a short time. An object recognition method in accordance with an exemplary aspect of the present invention is an object recognition method for recognizing a target object by using a preliminarily-created object model. The object recognition method generates a range image of an observed scene, detects interest points from the range image, extracts first features, the first features being features of an area containing the interest points, carries out a matching process between the first features and second features, the second features being features of an area in the range image of the object model, calculates a transformation matrix based on a result of the matching process, the transformation matrix being for projecting the second features on a coordinate system of the observed scene, and recognizes the target object with respect to the object model based on the transformation matrix.