Depth Image Motion Estimation via Optical Flow
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating the movement of an object in a static environment using optical recognition are either computationally expensive, unreliable, or require excessive processing resources, making them inadequate for real-time processing.
Innovation Solution
A method that processes video streams of depth images to estimate movement by exploiting temporal variations and spatial gradients, using mathematical formulations and numerical methods to approximate derivatives, allowing for efficient calculation of motion components such as linear and angular velocities through the use of specific parameters and linear system inversion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If robust point descriptors (SURF or SIFT type) are used for matching depth images, then matching reliability is improved, but computing resources and processing time increase significantly
Solution Approach 1:
The patent extracts and processes only the necessary depth information from depth images, rather than using complex robust point descriptors. By focusing on depth map analysis and optical flow computation specifically designed for depth data, the method achieves reliable matching without the heavy computational overhead of SURF or SIFT algorithms.
Solution Approach 2:
The patent replaces the mechanical/computational approach of feature descriptor matching with a physics-based optical flow model tailored for depth images. This substitution uses the continuity equation and depth-time derivative relationships to compute motion directly from depth variations, avoiding the need for complex feature extraction and matching pipelines.
2Ease of operation
If ICP algorithm is used for motion estimation, then initialization requirements are reduced, but computation time increases and real-time processing becomes difficult
Solution Approach 1:
The patent performs preliminary computation of depth-time derivatives and optical flow fields from consecutive depth images before solving the motion estimation problem. By pre-computing these intermediate quantities, the method avoids the iterative nature of ICP and enables direct, closed-form solution for motion parameters, achieving real-time processing speeds.
Solution Approach 2:
The patent computes optical flow and depth derivatives for all image pixels (excessive action), then uses this comprehensive data to directly determine motion parameters without iteration. This approach trades the computational expense of full-field computation for the efficiency of a non-iterative solution, enabling real-time performance.
3Measurement precision
If conventional RGB images are processed separately with robust point descriptors, then matching accuracy is improved, but device complexity and processing resources increase
Solution Approach 1:
The patent merges the processing of depth information and motion estimation into a unified framework. By combining depth map analysis with optical flow computation in a single integrated pipeline, the method achieves accurate motion measurement without requiring separate RGB image processing and feature matching systems, thereby reducing device complexity.
4Reliability
If iterative methods are used for depth image correlation, then matching robustness is improved, but computation time increases significantly
Solution Approach 1:
The patent replaces iterative correlation methods with a physics-based optical flow model that directly computes motion from depth-time derivatives. This substitution uses the continuity equation and kinematic relationships to provide a closed-form solution, eliminating iterative loops while maintaining robustness through the physical constraints embedded in the model.
Data Source
Figure 1
Figure 2a
Figure 2b
AI summary
The present invention relates to a method for estimating the movement of an object (3) moving about an environment (∑), the method being characterised in that it includes the steps of: (a) acquiring, by optical acquisition means (2) secured to the object (3), at least two consecutive depth images of the environment (∑); (b) selecting, by data-processing means (11) of a device (1), at least one sub-portion (Ωi) of the two depth images; (c) for each selected sub-portion (Ωi) of the two depth images, calculating by means of the data-processing means (11) on the basis of the depth values of the pixels in the sub-portion (Ωi): a first parameter (γi) representing a difference between the two depth images of a volume associated with a solid angle defined by a contour (∂Ωi) of the sub-portion (Ωi); a second parameter (αi) representing a volume vector quantity along said contour (∂Ωi); and a third parameter (βi) representing a surface vector quantity that results from said contour (∂Ωi); and (d) estimating, by the data-processing means (11), at least one component of the movement of said object (3) on the basis of the first, second and third parameters (αi, βi, γi) associated with each sub-portion (Ωi).