Detection device for detecting a useful signal in an acoustic input signal

Abstract

The invention relates to a detection device comprising a detector (D) configured to detect a useful signal in an acoustic input signal (ES) and configured to output a detection result (DE) upon detection of the useful signal as well as to output a quality signal (QS) indicating the reliability of the detection, wherein a quality signal limiter (QSL) is arranged and configured to use a false positive matrix (FPM) and a false negative matrix (FNM) to check whether the detection quality is limited by current states of an ego vehicle (ZEF) and current states of components (ZA) of the ego vehicle and, in this case, to reduce the quality signal (QS).

Description

[0001] The invention relates to a detection device for detecting a useful signal in an acoustic input signal according to the preamble of claim 1, an ego vehicle according to the preamble of claim 3 and a method for training a detection device for detecting a useful signal in an acoustic input signal according to the preamble of claim 5.

[0002] DE 10 2021 123 020 A1 describes a method for classifying a noise on board a vehicle, comprising steps of detecting a noise on board a vehicle; recognizing a predetermined pattern in the noise; and determining a classification associated with the pattern.

[0003] The invention is based on the objective of providing a novel detection device for detecting a useful signal in an acoustic input signal, a novel ego vehicle and a novel method for training a detection device for detecting a useful signal in an acoustic input signal.

[0004] The object is solved according to the invention by a detection device for detecting a useful signal in an acoustic input signal with the features of claim 1, by an ego vehicle with the features of claim 3 and by a method for training a detection device for detecting a useful signal in an acoustic input signal with the features of claim 5.

[0005] Advantageous embodiments of the invention are the subject of the dependent claims.

[0006] A detection device is proposed, comprising a detector configured to detect a useful signal in an acoustic input signal and configured to output a detection result upon detection of the useful signal, as well as to output a quality signal indicating the reliability of the detection. According to the invention, a quality signal limiter is arranged and configured to use a false-positive matrix and a false-negative matrix to check whether the current states of an ego vehicle and the current states of components of the ego vehicle, in which the detection device may be located, impair the detection quality, and in this case, to reduce the quality signal.

[0007] The false positive matrix can contain information about which states and / or combinations thereof in an input signal without a useful signal can be falsely detected by the detector, particularly with a certain probability.

[0008] The false negative matrix can contain information about which states and / or combinations thereof in an input signal with a superimposed useful signal may result in non-detection or poor detection of the useful signal, particularly with a certain probability.

[0009] The false positive matrix and the false negative matrix can be parts of a quality limitation look-up table that the quality signal limiter accesses.

[0010] In one embodiment, the useful signal is at least a special signal of a vehicle with priority.

[0011] According to one aspect of the present invention, an ego-vehicle is proposed comprising at least one external microphone for detecting an input signal, a detection device as described above, one or more components, and detectors for detecting the current states of the ego-vehicle and the current states of the component.

[0012] In one embodiment, the component comprises assemblies and non-assemblies. The assemblies essentially form rotating systems that generate noise, in particular a motor and / or an auxiliary unit, a transmission, an air conditioning compressor, a blower, a power steering system, a brake pressure generator for a brake booster, a pump, a compressor, a wheel bearing, a driveshaft, a wheel, and / or a tire. Each of the assemblies mentioned by way of example may be present multiple times. The assemblies may also form translational systems that generate noise, e.g., seat adjustment. The non-assemblies essentially form static systems that generate aerodynamic noise, in particular a vehicle body.

[0013] According to one aspect of the present invention, a method for training a detection device to detect a useful signal in an acoustic input signal, in particular the detection device described above, and especially in the ego-vehicle described above, is proposed. According to the invention, a plurality of states of the ego-vehicle and states of the component are traversed, and disturbances caused by wind and / or vehicles in the vicinity of the ego-vehicle and / or at least one component of the ego-vehicle are detected in the absence of a useful signal using at least one microphone arranged on the ego-vehicle as an input signal and stored in a database with an assignment to the states of the ego-vehicle and the states of the component.

[0014] In one embodiment, a power spectrum is further determined from the recorded associated input signal and stored in the database.

[0015] In one embodiment, a multitude of driving wind speeds, engine speeds and / or generator speeds and / or combinations thereof are set as states, and the corresponding power spectra from the acoustic input signal are recorded and stored in the database along with their assignment to the states.

[0016] In one embodiment, to generate a false-positive matrix, input signals for the various states and / or their combinations without a useful signal, stored in the database, are fed to the detector and analyzed by it, wherein, in the event of a false detection of a useful signal, the associated state of the input signals for this case is isolated and stored in the false-positive matrix.

[0017] In one embodiment, to generate a false negative matrix, input signals for the various states and / or their combinations with a superimposed useful signal, stored in the database, are fed to the detector and analyzed by it, wherein, in the event of non-recognition or poor recognition of the useful signal, the associated state of the input signals for this case is isolated and stored in the false negative matrix.

[0018] In one embodiment, a quality limitation look-up table is formed from the false positive matrix and the false negative matrix.

[0019] The present invention describes a method for reducing interference signals during the detection of emergency vehicles using external microphones. The interference signals are first detected and classified by the external microphones. A detector for a desired signal is designed without interference signals and then subjected to them. The detection result is then evaluated by creating various matrices (false positive, false negative) and generating a quality limitation lookup table, which is then applied during operation of the external microphone to improve the detection of emergency vehicles.

[0020] The solution according to the invention enables a more accurate and reliable quality signal for the acoustic detection of signals from emergency vehicles. Furthermore, false positive detections can be reduced.

[0021] Exemplary embodiments of the invention are explained in more detail below with reference to drawings.

[0022] This shows: Fig. 1. A schematic diagram of a first step in a design phase of a method for reducing interference signals in the detection of emergency vehicles using a detector. Fig. 2 a schematic diagram of a second step of the design phase, Fig. 3 a schematic diagram of a third step of the design phase, and Fig. 4 a schematic diagram of an application phase of the detector.

[0023] Corresponding parts are marked with the same reference symbols in all figures.

[0024] The present invention proposes a method for reducing interference signals during the detection of useful signals via external microphones. The interference signals are first detected and classified via the external microphones. The method can be carried out partially or completely in a vehicle, hereinafter also referred to as an ego-vehicle.

[0025] The external microphones are intended for the acoustic detection of special signals from vehicles with right-of-way, such as emergency vehicles. External microphones installed on the Ego vehicle can be affected or disrupted by aerodynamic disturbances, such as separating vortices and components like vehicle assemblies. These disturbances are not necessarily constant but can depend on the speed and type of vehicle, or on the design and condition of the external microphone component.

[0026] According to the present invention, disturbances are pre-determined in a design phase, and the resulting findings are taken into account in the design of the quality signal QS. If a false-positive detection occurs in the signal path due to the disturbances, this can be detected by the method described herein and communicated to a subsequent processing system by means of an additional quality signal QS.

[0027] Fig. Figure 1 is a schematic diagram of the first step of a design phase, also known as the learning phase or training phase, of the process.

[0028] In this process, disturbances caused by wind and / or vehicles in the vicinity and / or at least one component of the ego-vehicle are individually recorded by measurements taken directly at the ego-vehicle as an input signal ES, for example, using at least one or more microphones mounted on the ego-vehicle, in the absence of a signal such as a special signal from an emergency vehicle or a vehicle with right-of-way. The result of the measurement is a database DB, also known as a Vehicle and Component Induced Noise Signal Database, based on all states of the ego-vehicle and its components.The components can include assemblies that essentially form rotating systems which generate noise, in particular an engine and / or an auxiliary unit, a transmission, an air conditioning compressor, a blower, a power steering system, a brake pressure generator for a brake booster, a pump, a compressor, a wheel bearing, a driveshaft, a wheel and / or a tire. The components can also include non-assemblies that essentially form static systems which generate aerodynamic noise, such as a car body. During the design phase, a multitude of states of the ego vehicle ZEF and states of the component ZA are traversed, in particular all possible states, and a corresponding performance spectrum is determined from the recorded input signal ES and stored with the assignment in the database DB.For example, states ZEF, ZA are set to a variety or all possible wind speeds, engine speeds and / or generator speeds and / or combinations thereof, and the corresponding power spectra from the acoustic input signals ES are recorded and stored in the database DB with their assignment to the states ZEF, ZA.

[0029] Fig. Figure 2 is a schematic diagram of a second step in the design or learning phase. Using input signals ES, which consist solely of a useful signal without interference, a detector D for the useful signal with associated detection parameters DP, in particular at least one filter transfer function FÜF, is designed in a detector design step DES. The detector D for the useful signal is configured to recognize special signals from vehicles with right-of-way, especially emergency vehicles. Such detectors D and methods for their design are known in the prior art.

[0030] Fig. Figure 3 is a schematic diagram of a third step in the design phase or learning phase.

[0031] To generate a false positive matrix (FPM), interference signals stored in the database (DB) for the various states ZEF, ZA, and / or their combinations, without a corresponding signal, are fed to detector D and analyzed. If detector D falsely triggers at any point, this is considered a false positive. The corresponding status of the input signals ES for this case is isolated and stored in the false positive matrix (FPM). For example, if detector D falsely triggers at a specific motor speed, this is recorded in the false positive matrix (FPM). During subsequent operation, a corresponding limitation of the quality signal (QS) can then be implemented within this speed range. The false positive matrix (FPM) therefore records the cases in which no signal is present, but detector D falsely indicates that it has detected a signal.

[0032] Furthermore, to generate a false negative matrix (FNM), interference signals stored in the database (DB) for the various states ZEF, ZA, and / or their combinations, plus at least one useful signal, such as a special signal from a vehicle with right-of-way, particularly an emergency vehicle, are fed to detector D as input signal ES and analyzed by the detector. If the interference signals prevent or impair the detection of the useful signal, this is recorded in the false negative matrix (FNM). During subsequent operation, if such a state is reached (for example, due to interfering rotational speeds or wind speeds), the quality signal (QS) will also be limited. Thus, a subsequent system can recognize that the acoustic signal detector (D) cannot function reliably in this state.The false negative matrix FNM therefore stores the cases in which a useful signal is present but is not detected by detector D because, for example, it is at least partially masked by interference signals.

[0033] From the false positive matrix (FPM) and the false negative matrix (FNM), a quality limitation lookup table (QLLuT) is generated as the basis for a quality signal (QS), thus achieving false positive reduction. Using the knowledge stored in the QLLuT, the quality signal QS can be limited. This allows the system using detector D to be informed that the detection quality is limited in this situation.

[0034] The general quality signal QS is generated by the detector D of the useful signal. This detector D could, for example, be a detector D for detecting police or fire sirens. The detector D executes algorithms that provide the quality signal QS as metadata, which they use to communicate the reliability of the detection. For example, the detection reliability may be lower at a greater distance than at a shorter distance. Such algorithms can be misled by interference frequencies. An existing interference signal that exhibits some similarities to the useful signal could cause the algorithm to falsely detect the useful signal (for example, a police siren), for instance, with an 85% probability. However, the mechanism described in the present invention detects such possibilities of false detection during the training phase.

[0035] During operation, such a situation (the presence of an overriding interference signal, which could also be called a "phantom signal") is detected by the mechanism described here. In such a case, the algorithm's quality signal QS is reduced by this mechanism, for example, from 85% to a maximum of 35%, since an interference signal could cause a false positive. This quality signal QS can then be used, for example, in a subsequent fusion of a downstream system, such as a driver assistance system in a motor vehicle.

[0036] Fig. Figure 4 is a schematic diagram of an application phase of detector D.

[0037] An acoustic input signal ES, captured by at least one microphone, is fed to detector D. Detector D detects any useful signals present in the signal, such as a special signal from a vehicle with right-of-way, particularly an emergency vehicle, and forwards a detection result DE to a fusion machine FM of a subsequent system. Furthermore, detector D provides a quality signal QS as metadata, which indicates the probability of detection.

[0038] A quality signal limiter (QSL) uses the false positive matrix (FPM) and the false negative matrix (FNM) in the quality limitation look-up table (QLLuT) to check whether the current states of the ego vehicle (ZEF) and the current states of the ego vehicle's assemblies (ZA) impair the recognition quality. If so, the quality signal limiter (QSL) reduces the quality signal (QS) and forwards the reduced quality signal (QS) to the fusion machine (FM) of the following system.

[0039] This allows a detector D that has already been developed (for example, for other vehicles) to also be used on other vehicles. Reference symbol list D detector DB database DE Detection result DES detector design step DP detection parameters ES input signal FM Fusion Machine FNM False Negative Matrix FPM False Positive Matrix FÜF filter transfer function QLLuT Quality Limitation Look-up Table QS quality signal QSL Quality Signal Limiter ZA condition, condition of the components ZEF status, status of the Ego vehicle QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] DE 10 2021 123 020 A1

[0002]

Claims

[1] Detection device comprising a detector (D) configured to detect a useful signal in an acoustic input signal (ES) and configured to output a detection result (DE) upon detection of the useful signal and to output a quality signal (QS) indicating how reliable the detection is, characterized by a quality signal limiter (QSL) configured to use a false positive matrix (FPM) and a false negative matrix (FNM) to check whether the current states of an ego vehicle (ZEF) and current states of components (ZA) of the ego vehicle impair the detection quality and, if so, to reduce the quality signal (QS). [2] Detection device according to claim 1, characterized by that the signal is at least a special signal of a vehicle with priority. [3] Ego vehicle comprising at least one external microphone for detecting an input signal (ES), a detection device according to claim 1 or 2, components and detectors for detecting the current states of the ego vehicle (ZEF) and the current states of the components (ZA). [4] Ego vehicle according to claim 3, characterized by , that the components comprise assemblies and non-assemblies, wherein the assemblies essentially comprise rotating systems which generate noise, in particular an engine and / or an auxiliary unit, a transmission, an air conditioning compressor, a blower, a power steering system, a brake pressure generator for a brake booster, a pump, a compressor, a wheel bearing, a driveshaft, a wheel and / or a tire, and wherein the non-assemblies essentially comprise static systems which generate aerodynamic noise, in particular a body. [5] Method for training a detection device for detecting a useful signal in an acoustic input signal (ES), in particular the detection device according to claim 1 or 2, especially in an ego vehicle according to claim 3 or 4, characterized by , that a multitude of states of the ego vehicle (ZEF) and of states of the component (ZA) are traversed and disturbances caused by wind and / or vehicles in the vicinity of the ego vehicle and / or at least one component of the ego vehicle are detected in the absence of a useful signal, with at least one microphone arranged on the ego vehicle as an input signal (ES) and stored in a database (DB) with an assignment to the states of the ego vehicle (ZEF) and the states of the component (ZA). [6] Method according to claim 5, characterized by, that a power spectrum is determined from the recorded associated input signal (ES) and stored in the database (DB). [7] Method according to claim 5 or 6, characterized by , that a variety of driving wind speeds, engine speeds and / or generator speeds and / or combinations thereof are set as states (ZEF, ZA) and the corresponding power spectra from the acoustic input signal (ES) are recorded and stored in the database (DB) with their assignment to the states (ZEF, ZA). [8] Method according to any one of claims 5 to 7, characterized by, that to generate a false positive matrix (FPM) in the database (DB) input signals (ES) for the various states (ZEF, ZA) and / or their combinations without a useful signal are fed to the detector (D) and analyzed by it, wherein in the event of a false detection of a useful signal the associated state (ZEF, ZA) of the input signals (ES) is isolated for this case and stored in the false positive matrix (FPM). [9] Method according to any one of claims 5 to 8, characterized by , that to generate a false negative matrix (FNM) stored in the database (DB) input signals (ES) for the various states (ZEF, ZA) and / or their combinations with superimposed useful signal are supplied to the detector (D) and analyzed by it, wherein in case of non-recognition or poor recognition of the useful signal the associated state (ZEF, ZA) of the input signals (ES) is isolated for this case and stored in the false negative matrix (FNM). [10] Method according to claims 8 and 9, characterized by , that a Quality Limitation Look-up Table (QLLuT) is formed from the false positive matrix (FPM) and the false negative matrix (FNM).

Images