3D Point Cloud Extraction Using Robust Feature Point Denoising
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 3D point cloud generation methods require specialized and costly hardware, leading to noise and errors, and non-specialized hardware generates even noisier point clouds, necessitating effective noise removal and extraction of relevant object data.
Innovation Solution
A computer-implemented method using imaging mobile devices to capture images from different viewpoints, identify robust feature points, generate median feature points, and apply de-noising phases to filter out noise and errors in the 3D point cloud.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If specialized hardware (3D laser scanners) is used to generate 3D point clouds, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent uses 2D images captured by ordinary cameras as copies or representations of the 3D scene, processing these image copies through computer vision algorithms to reconstruct 3D point clouds. This avoids the need for complex specialized scanning hardware while achieving comparable measurement precision through computational methods rather than physical scanning devices
Solution Approach 2:
The patent replaces mechanical 3D scanning systems with a computational approach using 2D images. Instead of using physical laser scanners or structured light systems to directly measure 3D geometry, the system uses image processing and stereo vision algorithms to derive 3D information from 2D photographs, substituting mechanical measurement systems with computational reconstruction methods
2Device complexity
If non-specialized hardware (mobile imaging devices) is used to generate 3D point clouds, then device complexity is reduced, but measurement precision deteriorates due to noise and errors
Solution Approach 1:
The patent performs preliminary actions by capturing multiple 2D images from different viewpoints before generating the final 3D point cloud. These preliminary images serve as raw data that undergoes extensive processing including feature point detection, stereo matching, and iterative optimization to remove noise and errors, thereby improving measurement precision despite using simple mobile devices
Solution Approach 2:
The patent implements feedback mechanisms through iterative optimization processes where the 3D point cloud is repeatedly adjusted and refined based on the 2D image data. The system uses feedback from image features and depth information to correct errors and reduce noise in the point cloud, continuously improving measurement precision until convergence is achieved
3Adaptability or versatility
If mobile imaging devices move around the object to capture images, then adaptability is improved, but measurement precision deteriorates due to accumulation of errors causing drift
Solution Approach 1:
The patent segments the 3D reconstruction process into independent modules: image capture, feature point detection, stereo matching, and optimization. By processing images and features independently before integrating them into the final point cloud, the system reduces the propagation of errors through the pipeline, maintaining measurement precision despite the flexible mobile capture approach
Data Source
AI summary
The disclosed systems, structures, and methods are directed to generating data points in a 3D point cloud generated from a plurality of images of an object, each image having been captured by an imaging mobile device from a corresponding point of view. The method comprises identifying, for identified feature points of the object, a list of 3D coordinates, and a list of 2D coordinates, an entry of said list being 2D coordinates of a projection of the feature data point on the corresponding image. In response to determining that, for a given feature point, a number of images on which the corresponding projected 2D coordinates fall onto a pixel range of said images is above a pre-determined threshold, the given feature point is marked as a robust feature point. A median feature point is generated from the list of 3D coordinates of the robust feature point.