3D Orientation Calculation Using Voxel Grids and Integral Volumes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for calculating 3D orientations in real-time scene analysis are limited by precision, cost, and computing resources, particularly due to reliance on 2D supports that distort 3D neighborhood information and require high-cost equipment.
Innovation Solution
A method that calculates 3D orientations by quantifying points in a 3D grid of voxels, computing statistical characteristics, and using integral volumes to determine the orientation from a consistent 3D neighborhood, allowing for exact and efficient orientation calculation on standard architectures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If 2D supports are used to calculate 3D orientations in real-time, then computation speed is improved, but measurement precision deteriorates due to distortion of 3D neighborhood information
Solution Approach 1:
The patent transitions from 2D image space to 3D voxel space by organizing 3D points into a voxel grid. This dimensional change allows the system to preserve and process true 3D neighborhood relationships while maintaining computational efficiency through the structured grid organization, resolving the contradiction between using 2D supports for speed and 3D data for precision.
2Measurement precision
If high-cost 3D sensors are used to improve depth image precision, then measurement precision is improved, but device cost increases
Solution Approach 1:
The patent creates a computational model (voxel grid) that replicates and processes 3D spatial relationships from 2D input data. Instead of requiring expensive dedicated 3D sensors, the system uses computational geometry to construct and process 3D representations from standard camera inputs, achieving 3D processing capabilities without high-cost hardware.
3Measurement precision
If 3D voxel grid with integral volumes is used to calculate orientations, then measurement precision is improved, but computing resources increase
Solution Approach 1:
The patent pre-computes and stores integral volumes for each voxel in the grid before orientation calculation. By preparing these cumulative sums in advance, the system avoids performing repeated summation operations during the actual orientation computation, significantly reducing the computational resources required at runtime while maintaining precise 3D neighborhood analysis.
Data Source
Figure 1~3
Figure 2
AI summary
The invention relates to a method for characterising a scene by calculating the 3D orientation of observed elements of the scene, which comprises a step of calculating 3D points of the scene (step B1), characterised in that it also comprises the following steps: regularly quantifying the 3D points of the scene, according to three axes, in a predetermined 3D grid of voxels (step B2); for each non-empty voxel of the 3D grid, calculating N predefined statistical features from the coordinates of points contained within the voxel, with N an integer greater than 1, and for each of the N features, defining a 3D grid associated with said feature (step C); for each 3D grid associated with a statistical feature, calculating an integral 3D grid associated with said statistical feature, each voxel of the integral 3D grid including an integral of said statistical feature (step D); for each non-empty voxel of the 3D grid of voxels: defining a 3D rectangular parallelepipedal vicinity centred on said voxel and having predefined 3D dimensions, and eight vertices of said vicinity; for each of the N statistical features, calculating the integral of the statistical feature in said vicinity from the eight vertices considered in the integral 3D grid associated with said statistical feature; calculating the 3x3 covariance matrix of the 3D points in said vicinity; and calculating the orientation of said voxel from the 3x3 covariance matrix (step E).