3D Point Cloud Segmentation for Mobile Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current 3D modeling techniques using photogrammetry generate sparse and cluttered 3D point clouds with significant background and foreground noise, making processing complex and computationally expensive, especially when utilizing Bundle Adjustment and local plane fitting.
Innovation Solution
A user-assisted segmentation method that uses gestural inputs, such as touch gestures, to isolate sections of a 3D point cloud corresponding to a target object, employing kernels like triangular or Gaussian kernels to prune the point cloud and reduce complexity, allowing for efficient generation of 3D models on mobile devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If photogrammetry is used to generate 3D point clouds from multiple images, then 3D models can be created, but the point clouds become sparse and cluttered with significant background and foreground noise
Solution Approach 1:
The patent applies segmentation by dividing the 3D point cloud into multiple sections based on depth information and user-selected seed points. The point cloud is segmented into foreground objects and background elements, allowing selective processing of only relevant regions. This reduces the overall complexity by focusing computational resources on segmented portions rather than processing the entire cluttered point cloud.
Solution Approach 2:
The patent extracts the target object from the cluttered point cloud by using depth maps and seed points to isolate the object of interest. The extraction process removes background and foreground noise, keeping only the relevant 3D points that constitute the target object. This significantly reduces point cloud complexity while maintaining modeling accuracy.
2Manufacturing precision
If Bundle Adjustment and local plane fitting are applied to process the entire 3D point cloud, then 3D model quality improves, but processing becomes computationally expensive
Solution Approach 1:
The patent segments the point cloud into multiple sections using depth information and seed points before applying Bundle Adjustment and local plane fitting. By processing only the segmented portions containing the target object rather than the entire point cloud, the computational energy required for these intensive operations is significantly reduced while maintaining model precision.
Solution Approach 2:
The patent applies local plane fitting and Bundle Adjustment only to specific segmented regions of the point cloud that contain the target object, rather than uniformly processing the entire point cloud. This localized approach maintains high precision for the object of interest while reducing overall computational energy consumption by excluding background and irrelevant foreground elements.
3Area of stationary object
If the entire 3D point cloud is processed for 3D model generation, then complete scene coverage is achieved, but processing time and computational resources increase
Solution Approach 1:
The patent segments the 3D point cloud into multiple sections based on depth maps and user-selected seed points, identifying regions containing the target object versus background areas. By processing only the relevant segmented portions, the patent reduces processing time significantly while maintaining adequate scene coverage for the object of interest.
Solution Approach 2:
The patent applies partial action by processing only the necessary portions of the point cloud that contain the target object, rather than processing the entire point cloud. This selective processing approach uses just enough computational resources to achieve the modeling goal without the excessive time and resource consumption of complete scene processing.
Data Source
AI summary
Techniques for segmentation of three-dimensional (3D) point clouds are described herein. An example of a method for user-assisted segmentation of a 3D point cloud described herein includes obtaining a 3D point cloud of a scene containing a target object; receiving a seed input indicative of a location of the target object within the scene; and generating a segmented point cloud corresponding to the target object by pruning the 3D point cloud based on the seed input.