Acoustic Event Detection via Spatial-Temporal Impulse Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing acoustic signal detection systems struggle in environments with unknown noise sources or large noise regions, as they often fail to filter noise effectively and may remove relevant signals, especially when the noise position is not well-defined.
Innovation Solution
A method involving a computing unit that processes acoustic signal data from sensors by aligning it to a common time reference, identifying candidate impulses, determining signal source times, and generating an indication of acoustic events based on weighted sums within spatial and temporal grids, allowing for effective detection of events even in challenging environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If filter based solutions are used to attenuate noise at pre-defined frequencies, then noise filtering capability is improved, but relevant signals may be removed from the obtained signal data
Solution Approach 1:
The patent changes the fundamental parameter of noise filtering from frequency-domain attenuation to time-domain impulse identification and localization. Instead of filtering based on pre-defined frequencies, the system identifies candidate impulses, determines their source times, and localizes them in space and time to distinguish noise from relevant signals without removing useful information
Solution Approach 2:
The patent replaces the traditional filter-based mechanical system with an impulse-based detection system. Rather than using filters to attenuate signals, the system uses impulse identification, time determination, and spatial localization to differentiate between noise and relevant acoustic events, substituting a more sophisticated detection mechanism
2Measurement precision
If microphone arrays are focused onto a volume of space by scaling and delaying signals, then signal concentration from focal volume is improved, but signals from elsewhere may cancel out and relevant events outside focal volume are lost
Solution Approach 1:
The patent segments the monitoring space into multiple sub-spaces and independently identifies candidate impulses in each sub-space. This allows the system to maintain focused detection accuracy in each local region while collectively covering the entire space, avoiding the need to choose between focal concentration and broad coverage
Solution Approach 2:
The patent adds the spatial dimension to impulse identification by determining sensor coordinates and localizing impulses in specific sub-spaces. This dimensional approach allows simultaneous maintenance of local detection precision and global coverage by treating different spatial regions as independent detection zones
3Object-affected harmful factors
If filter based solutions operate in the frequency domain to attenuate signals at pre-defined frequencies, then noise attenuation is improved, but the solution is not applicable when noise source position is not well known or noise is generated over a large region
Solution Approach 1:
The patent enables the system to automatically identify and localize noise sources without requiring pre-defined frequency information or known noise source positions. By identifying candidate impulses and determining their source times and sensor coordinates, the system adapts to any noise environment self-service style, making filter-based frequency domain solutions obsolete
Solution Approach 2:
The patent fundamentally changes the parameters used for noise handling from frequency-based attenuation to time-based impulse identification and space-based localization. This parameter transformation allows the system to handle any noise source configuration, whether point-source or distributed, without requiring prior knowledge of noise characteristics
Data Source
AI summary
Disclosed is a method for detecting an acoustic event of interest in a space. In the method acoustic signal data is obtained from sensors and at least some candidate impulses are determined. The candidate impulses are mapped to a representation on a basis of an origin of the candidate impulse in question and it is determined, from the generated representation, at least one indication quantity representing a likelihood of an acoustic event of interest taking place in the specified positions in space and time. Finally, the at least one indication quantity is compared to a predetermined threshold and an indication is generated if the at least indication quantity meets the predetermined threshold. Also disclosed is a computing unit and a computer program product.