Vision-Based Tactile Sensing Using Marker Displacement Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current tactile sensors require a large number of training samples (up to 20,000) for effective operation, and their algorithms for curvature estimation are discontinuous and inefficient.
Innovation Solution
A vision-based tactile measurement method using a feedforward neural network that processes the displacement of marking points on a flexible sensing face, reducing the number of required training samples and improving training efficiency by simplifying the input features to displacement and deformation, while enabling continuous curvature estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a convolutional neural network (CNN) is used for tactile measurement, then the measurement capability is improved, but the training complexity increases and requires up to 20,000 training samples
Solution Approach 1:
The patent extracts only the essential feature (displacement of marking points) from the complete image data, transforming the complex image processing task into a simpler displacement measurement task. This extraction principle reduces the training complexity while maintaining measurement accuracy by focusing only on the critical information needed for tactile measurement.
Solution Approach 2:
The patent replaces the complex CNN mechanical system with a simpler feedforward neural network that processes pre-calculated displacement features. This substitution reduces the number of training samples needed from 20,000 to a much smaller set, while maintaining the ability to perform accurate tactile measurements including curvature estimation.
2Measurement precision
If a convolutional neural network (CNN) is used for tactile measurement, then the measurement capability is improved, but the number of training samples required increases to up to 20,000
Solution Approach 1:
The patent extracts only the essential feature (displacement of marking points) from the complete image data, transforming the complex image processing task into a simpler displacement measurement task. This extraction principle reduces the training complexity while maintaining measurement accuracy by focusing only on the critical information needed for tactile measurement.
Solution Approach 2:
The patent replaces the complex CNN mechanical system with a simpler feedforward neural network that processes pre-calculated displacement features. This substitution reduces the number of training samples needed from 20,000 to a much smaller set, while maintaining the ability to perform accurate tactile measurements including curvature estimation.
3Ease of operation
If traditional algorithms are used for curvature estimation, then the implementation is simple, but the estimation is discontinuous and inefficient
Solution Approach 1:
The patent replaces traditional discontinuous curvature estimation algorithms with a feedforward neural network that processes displacement features to provide continuous curvature estimation. This substitution maintains implementation simplicity while dramatically improving estimation efficiency and continuity, enabling real-time tactile measurement.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method achieves accurate tactile measurements with reduced training samples and continuous curvature estimation, simplifying the measurement process and enhancing the performance of tactile sensors.
Implementation Method 1
When an outer surface of the flexible sensing face is in contact with an object, the flexible sensing face is deformed, which causes locations of the multiple marking points on the inner surface to change due to the deformation
Implementation Method 2
The image sensing component collects an image of the inner surface of the flexible sensing face
Data Source
Figure 1~3
Figure 4~5
Figure 6~8
AI summary
A vision-based tactile measurement method; the method is executed by a chip, and the chip is connected to a tactile sensor; the tactile sensor comprises a sensing surface and an image sensing assembly, and the sensing surface is provided with a marking pattern; the method comprises: acquiring an image sequence collected by the image sensing assembly for the sensing surface, images in the image sequence comprising marking patterns; according to the marking patterns within adjacent images in the image sequence, calculating a difference feature of the marking patterns; and calling a feedforward neural network to process the difference feature of the marking patterns to obtain a tactile measurement result.