Range Image Plane Estimation via Recursive Triangle Expansion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating planes in range images struggle to accurately separate the foot of a human body from the floor surface, especially when the contact point is included in the captured image, and are prone to false recognition of smooth curves as planes, with limited ability to detect small steps or surfaces.
Innovation Solution
A method that converts 3D data into a houndstooth arrangement, computes normal vectors for small triangles, and recursively searches for plane candidates by comparing cross product magnitudes, allowing for accurate separation of planes with minimal computation, preventing false recognition of smooth curves and detecting small steps as planes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional range image processing is used to separate foot from floor, then computation is simplified by excluding contact areas, but the viewing angle must be intentionally restricted and contact points cannot be included in the captured image
Solution Approach 1:
The patent segments the range image into multiple local regions and applies plane estimation to each region independently. This allows the system to process the entire field of view including contact points between foot and floor, while maintaining computational efficiency through localized processing rather than global analysis.
Solution Approach 2:
The patent applies different processing strategies to different regions of the image. Plane estimation is performed on local regions where appropriate, while other areas may use different methods. This local quality approach enables flexible handling of contact points and varied surface types without requiring uniform processing of the entire image.
2Device complexity
If plane estimation is performed on single-pixel range data, then device complexity is reduced, but measurement precision deteriorates due to unstable range values
Solution Approach 1:
The patent merges range data from multiple adjacent pixels to create a stable plane estimation. By combining information from neighboring pixels within local regions, the system achieves reliable plane detection even when individual pixel range values are unstable, without requiring complex additional hardware.
Solution Approach 2:
The patent transitions from analyzing single-pixel range values to analyzing planar relationships across multiple pixels. By estimating planes through spatial relationships in the range image rather than relying on individual pixel stability, the system achieves precise measurement without increased device complexity.
3Measurement precision
If difference images from multiple range images are used to separate target objects, then separation accuracy improves, but the method only separates background passively and does not actively recognize planes
Solution Approach 1:
The patent performs preliminary plane estimation on the range image before target object separation. By actively recognizing planes in advance, the system can use this plane information to guide subsequent target separation processes, improving overall accuracy while maintaining automation throughout the workflow.
Solution Approach 2:
The patent uses estimated plane information as feedback to improve target object separation. The plane estimation results are fed back into the separation process to enhance accuracy, creating an automated system that actively recognizes planes and uses this information to improve overall performance rather than merely passively separating backgrounds.
4Measurement precision
If stereo images are used for plane estimation, then plane estimation accuracy improves, but device complexity increases and the method is limited to stereo image configurations
Solution Approach 1:
The patent extracts plane information directly from single range images without requiring stereo image pairs. By taking out the essential plane estimation function from the complex stereo imaging system, the method achieves accurate plane detection using only the range data already available from time-of-flight cameras, reducing device complexity while maintaining accuracy.
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
Enables clear separation of the foot from the floor surface with reduced computation, improving the accuracy of human body detection and preventing false plane recognition, while accurately identifying planes even with small steps or surfaces.
Implementation Method 1
an imaging element that acquires range information for pixel values and generates a range image based on a measurement of the time taken for projected light to be reflected back by a target
Data Source
AI summary
3D data of pixels arranged in a square lattice is converted to a houndstooth arrangement, a largest value among respective magnitudes of cross products of all pairings of the normal vector of a small triangle of interest and the normal vectors of three adjacent small triangles is derived, and a large triangle composed of the small triangle of interest and the adjacent small triangles is determined to be a plane candidate if the largest value is less than or equal to a first threshold, after which similar processing is repeated with this large triangle taken as a small triangle, and this repetition is recursively performed while enlarging the small triangle of interest, and when the area of a new large triangle is greater than or equal to a predetermined area, the region thereof is determined to be a plane and separated from a range image.