3D Object Model Fitting Using Segmented Point Cloud Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for fitting object models to three-dimensional point cloud data, especially when errors are large or occlusions occur, often result in significant deviations from the actual surface position and orientation, particularly when using image data-generated point clouds.
Innovation Solution
A computer program product that includes an image acquisition unit, a point cloud acquisition unit, a coordinate system specifying unit, an area detection unit, an extraction unit, a model acquisition unit, and a generation unit, which acquires image data, specifies a reference coordinate system, detects a target area, extracts relevant point cloud data, and generates target object information by fitting a target object model to the point cloud data under conditions that match the defined orientation of the object, ensuring accurate positioning and orientation of the model in a three-dimensional space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional least squares method or principal component analysis is used to fit a plane model to three-dimensional point cloud data, then the modeling process is simple and fast, but the position and orientation of the calculated surface deviate from the actual surface when the point cloud data has large errors or occlusions occur
Solution Approach 1:
The patent segments the three-dimensional point cloud data into multiple groups based on distance from the median value. By dividing the data into different segments and assigning different weights to each segment, the method can focus on the most reliable data points (those closest to the median) while reducing the influence of outliers and occluded points, thereby improving the accuracy and robustness of the surface fitting.
Solution Approach 2:
The patent applies different quality weights to different regions of the point cloud data based on their distance from the median value. Points closer to the median are assigned higher weights, indicating better quality and reliability, while points farther away are assigned lower weights. This local quality differentiation allows the fitting algorithm to prioritize reliable data points and minimize the impact of erroneous or occluded points on the final surface model.
2Manufacturing precision
If robust estimation or reduced contribution ratio techniques are used to reduce deviation, then the accuracy improves slightly, but the deviation remains large when a large number of point clouds of other objects are included
Solution Approach 1:
The patent implements segmentation by grouping point cloud data according to their distance from the median value, creating distinct segments that can be processed differently. This segmentation approach systematically separates reliable data from unreliable data, providing a structured method to handle mixed-quality point clouds without requiring complex iterative robust estimation algorithms.
Solution Approach 2:
The patent changes the parameter of contribution ratio dynamically based on the distance of each point from the median value. By adjusting the weight parameter according to distance, the method adaptively optimizes the fitting process, giving more importance to reliable points and less to unreliable points, thereby achieving better accuracy with a relatively simple weighted least squares approach.
3Ease of manufacture
If three-dimensional point cloud data is generated from image data using pixel correlation or neural networks, then the data acquisition is flexible and cost-effective, but the error in the point cloud data is larger compared to laser or radar sensors
Solution Approach 1:
The patent applies local quality weighting to compensate for the inherently lower precision of image-based point cloud data. By assigning higher weights to points closer to the median value and lower weights to points farther away, the method maximizes the utility of the available image-based data while minimizing the impact of errors, thereby achieving reasonable accuracy without requiring expensive laser or radar sensors.
Solution Approach 2:
The patent changes the contribution ratio parameter based on the statistical distribution of the point cloud data, specifically using the distance from the median value as a criterion. This parameter adjustment transforms the fitting process to be more tolerant of the larger errors inherent in image-based point cloud generation, allowing flexible and cost-effective data acquisition while maintaining acceptable modeling accuracy.
Data Source
AI summary
A computer program product includes a computer-readable recording medium on which programmed instructions causing a computer to perform the following processing are recorded. The computer specifies a reference coordinate system of a three-dimensional position. The computer detects, from an image data, an area in which a designated target object is included. The computer extracts, from three-dimensional point cloud data, extraction point cloud data representing a three-dimensional position of an object in a target area. The computer generates target object information. A shape and an orientation of a first portion of the designated target object are defined. The orientation is defined for one of coordinate axes in the reference coordinate system. The computer generates the target object information by fitting the target object model to the extraction point cloud data under a condition that an orientation of the target object model matches an orientation defined for the first portion.