Clothing-Aware 3D Torso Shape Generation for Accurate HRTFs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for generating 3D models for head-related transfer functions (HRTFs) are prone to distortion due to acoustically transparent surfaces like clothing, requiring manual processing that is time-consuming and error-prone.
Innovation Solution
A computer-implemented method using a machine learning model that processes 3D scans to generate clean torso shapes by removing occluding surfaces, trained on target shapes with heuristic constraints derived from anatomical relationships.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 3D scans include clothing surfaces, then the scan captures the complete external shape, but the HRTF synthesis becomes distorted due to acoustically transparent surfaces
Solution Approach 1:
The patent extracts and removes clothing surfaces from the 3D scan data, separating the acoustically relevant torso geometry from the acoustically transparent clothing layers. This extraction process isolates the harmful clothing surfaces and eliminates their distorting effect on HRTF synthesis while preserving the underlying torso shape.
Solution Approach 2:
The patent introduces an intermediary processing step between 3D scanning and HRTF synthesis that identifies and removes clothing surfaces. This intermediary layer acts as a filter that separates relevant from irrelevant geometric data, allowing the system to automatically eliminate clothing artifacts without requiring manual intervention or specialized scanning procedures.
2Measurement precision
If manual removal of clothing surfaces is performed, then HRTF synthesis accuracy improves, but the process becomes time-consuming and error-prone
Solution Approach 1:
The patent implements a self-service system where the processing pipeline automatically identifies and removes clothing surfaces without human intervention. The system uses the 3D scan data itself to train and execute the removal process, making the entire workflow autonomous and eliminating the need for manual artifact removal while maintaining high accuracy.
Solution Approach 2:
The patent replaces the manual mechanical process of artifact removal with an automated computational system. Instead of requiring operators to manually identify and remove clothing surfaces, the system uses machine learning models trained on 3D scan data to automatically perform the removal, significantly reducing time and eliminating human error.
3Ease of operation
If subjects wear clothing during 3D scanning, then the scan is practical and comfortable, but the resulting HRTF model contains distortion from clothing surfaces
Solution Approach 1:
The patent performs preliminary processing of the 3D scan data to remove clothing surfaces after scanning but before HRTF synthesis. This preliminary action allows subjects to wear comfortable clothing during scanning while ensuring that the clothing artifacts are automatically eliminated in the subsequent processing stage, preserving both scanning convenience and model accuracy.
Solution Approach 2:
The patent converts the harmful effect of clothing surfaces into a beneficial workflow feature. Instead of requiring subjects to remove clothing (which creates practical problems), the system embraces the clothing-covered scan as valid input and uses automated processing to extract the underlying torso geometry, turning a potential source of error into an opportunity for robust, automated artifact removal.
Data Source
AI summary
Disclosed embodiments include techniques for generating a clean 3D shape of a portion of the human torso of a subject from a 3D scan that includes overlaying surfaces that occlude a portion of the torso, where the overlaying surfaces are acoustically transparent, such as wrinkles on a t-shirt and/or other clothing surfaces. These techniques use heuristic constraints to generate a clean 3D shape of a human torso for the purpose of generating a more realistic head-related transfer function (HRTF) that is individualized for the subject. The improved HRTF, when applied to a sound system, leads to a more immersive and realistic acoustic experience for the subject when listening to music, experiencing a virtual reality experience, and/or the like.