Wearable Pattern for Voice Source Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for localizing the position of a voice signal source in noisy environments, such as hospitals and industrial settings, face challenges in accurately identifying the speaker when their face is covered, as audio localization techniques fail and computer vision methods have high false acceptance rates.
Innovation Solution
A system that uses a visually detectable pattern on a wearable object, such as a mask or badge, which is detected by a camera and processed to determine the position of the voice source, allowing for more reliable localization and adaptation of microphones for improved speech recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If audio localization techniques are used to track sound sources, then the system can identify speakers in noisy environments, but the technique fails when the face or part thereof is covered
Solution Approach 1:
The patent introduces a visually detectable pattern on wearable objects as an intermediary element. Instead of directly localizing sound sources or detecting faces, the system uses the camera to detect the visual pattern on the mask or badge, which serves as a mediator to determine the position of the voice source. This intermediary approach bypasses the limitations of both audio localization (which fails with covered faces) and face detection (which cannot see through masks).
Solution Approach 2:
The patent replaces the acoustic field-based localization method with an optical field-based method. Instead of using microphones and audio processing to locate speakers, the system uses a camera to capture visual patterns and determines positions through image processing. This substitution of the detection domain (from acoustic to optical) resolves the contradiction by making face coverage irrelevant to the detection process.
2Measurement precision
If computer vision techniques are used to identify persons, then the system can detect face positions, but the method has high false acceptance rate when faces are covered
Solution Approach 1:
The patent extracts the essential identification feature from the face itself and places it on the wearable object (mask or badge). Instead of relying on facial features that are obscured by masks, the system uses a visually detectable pattern that is deliberately placed on the object worn by the person. This extraction of the identification marker from the hidden facial region resolves the contradiction between detection precision and reliability.
Solution Approach 2:
The patent creates a visual copy or representation of the person's presence through the detectable pattern on their wearable object. Rather than attempting to see through or around the mask to detect the actual face, the system uses a deliberate visual marker that copies the positional information of the person in a detectable form. This copying approach eliminates the false acceptance problems associated with trying to detect covered faces.
3Ease of operation
If face localization methods are used for position determination, then the system can identify speakers, but the method becomes unreliable when the face is covered by protective equipment
Solution Approach 1:
The patent applies preliminary action by placing the visually detectable pattern on the wearable object before the person enters the detection zone. The pattern is pre-positioned on the mask or badge, so when the person approaches, the camera can immediately detect the pattern and determine position without needing to perform complex face localization. This preliminary placement of the detection marker simplifies the operation and maintains reliability regardless of face coverage.
Data Source
AI summary
The invention relates to localizing the position of a person speaking by using pictures of a pattern (21) on an object (20) worn by the person. The object (20) carries a complex pattern (21) that is optimized for determining the orientation of the object (20), the distance from the object to a microphone device (14) and/or to a camera (11). Moreover, the pattern (21) may be arranged for identifying the person carrying the object (20). The determination of the position of the person carrying the object (20) may be used to enhance speech recognition (SR) and/or to provide hands-free voice control of devices (DC), e.g. in hospitals or in industrial settings.