3D Point Cloud Filtering for Cell Tower Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generating a 3D model of a cell tower using drone images is challenging due to the large number of points captured in the point cloud, making it difficult to automatically extract points belonging to the cell tower from surrounding background objects and ground, especially when using low oblique imagery.
Innovation Solution
A method and apparatus that utilize bounding box level visual object detection on drone images with known camera parameters to project points from a point cloud onto image planes, filter background and ground points, and generate a reduced point cloud containing only relevant points of the cell tower by analyzing the frequency of point inclusion in bounding boxes across multiple images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a 360° orbit with consecutive images is used to generate a 3D point cloud, then complete coverage of the cell tower is achieved, but the point cloud contains tens of millions of points including background and ground, making automatic extraction difficult
Solution Approach 1:
The patent segments the large point cloud by defining multiple areas of interest (AOIs) corresponding to different images in the image set. Each AOI is associated with specific image regions, and points are classified based on which AOIs they fall into. This segmentation approach divides the complex task of extracting cell tower points from the entire point cloud into smaller, more manageable subsets, improving both efficiency and accuracy.
Solution Approach 2:
The patent performs preliminary action by defining areas of interest and establishing correspondence between image regions and point cloud regions before the actual extraction process. The AOIs are pre-defined based on image content, and the mapping between image pixels and 3D points is established in advance, allowing for more efficient and accurate point cloud filtering during the extraction phase.
2Reliability
If low oblique imagery with down tilt of 45-60° is used, then stable ground landmarks are available for 3D modeling, but the point cloud captures large parts of surrounding scene including background objects and ground
Solution Approach 1:
The patent extracts only the relevant points belonging to the cell tower from the comprehensive point cloud by using area of interest definitions derived from image analysis. The system identifies and extracts cell tower points while excluding background and ground points, transforming the complex full-scene point cloud into a focused subset containing only the target object points.
Solution Approach 2:
The patent introduces an additional dimension of analysis by mapping 2D image regions to 3D point cloud regions through areas of interest. This dimensional bridge allows the system to use 2D image information (which is easier to analyze) to guide 3D point cloud filtering, effectively using another dimension to solve the complexity problem.
3Loss of information
If tens of millions of points are captured in the point cloud, then complete scene information is obtained, but automatic extraction of cell tower points becomes challenging
Solution Approach 1:
The patent introduces areas of interest as an intermediary between the image data and the point cloud data. The AOIs serve as a mediator that translates 2D image regions into 3D point cloud regions, enabling the system to leverage image-based object detection to guide point cloud extraction without directly processing the entire point cloud.
Solution Approach 2:
The system performs preliminary analysis on the image set to define areas of interest before attempting point cloud extraction. This preliminary action identifies potential cell tower regions in 2D images, which then guide the subsequent 3D point cloud filtering process, making the extraction task significantly easier and more accurate.
Data Source
AI summary
A method (1200) for producing a reduced point cloud comprising an object of interest (OOI) from an original point cloud comprising the OOI. The method includes obtaining (s1202) the original point cloud; obtaining (s1204) a set of N images, each of the N images comprising an image of the OOI from a unique position relative to the OOI; for each one of the N images, defining (s1206) an area of interest in the image that obtaining a set of N images, each of the N images that includes the OOI, thereby defining N areas of interest; for each point included in the original point cloud, determining (s1208), for each one of the N areas of interest, whether the point is located in the area of interest; for each point included in the original point cloud, determining (s1210) a first metric for the point based on the total number of the N areas of interest in which the point is determined to be located; and for each point included in the set of points, determining (s1212) whether or not to include the point in the reduced point cloud based on the first metric for the point.