DE102010003317B4 - Method and device for combining sensor data for a classification of a collision of a vehicle - Google Patents

Abstract

A method for combining sensor data for a classification of a collision of a vehicle (100), comprising the following steps:
Receiving (201) time traces of sensor signals (531, 535) via an interface, the sensor signals representing signals provided by different sensors (102, 104, 106; 402, 406, 407, 408); and
Combining (203) the time histories of the sensor signals to obtain a time course of a combined signal (654) for classifying the collision,
characterized in that
a first set of sensor signals (531, 535) of a first physical quantity and a second set of sensor signals of a second physical quantity are assigned, and a step of normalizing at least one of the sets of the sensor signals to the sensor signals of the first set and the second set prior to the step of combining (203).

Description

State of the art

The present invention relates to a method for combining sensor data for a classification of a collision of a vehicle, to a corresponding device and to a corresponding computer program product.

The DE 10 2007 027 649 A1 describes a method for controlling personal protection devices, in which a feature vector having at least two features is formed from a signal of an accident sensor system. Personal protection means are activated as a function of the feature vector by means of a core algorithm. The feature vector is classified by a support vector machine and the core algorithm is influenced by this classification.

The DE 10 2007 048 884 A1 deals with a method for controlling personal protective equipment in a side impact for a vehicle. A crash type is determined by means of two signals from differently installed sensors.

The DE 10 2007 035 511 A1 is concerned with a method for triggering protective measures to protect the occupants of a motor vehicle, in which signals are provided by movement sensors arranged on the motor vehicle, by means of which a collision of the motor vehicle with an obstacle can be detected.

The DE 10 2007 006 771 A1 deals with a method for controlling personal protective equipment in the event of a side impact. The side impact is detected by spaced-apart impact sensors. Based on a signal difference between the signals of the impact sensors, the activation case is distinguished from the non-activation case.

Disclosure of the invention

Against this background, the present invention proposes a method for combining sensor data for a classification of a collision of a vehicle, furthermore a device which uses this method and finally a corresponding computer program product according to the independent patent claims. Advantageous embodiments emerge from the respective subclaims and the following description.

The invention is based on the recognition that a generation of a control channel for machine-based learning methods is possible by means of a combination of sensor data. Such a machine-based learning method (Support Vector Machine) can be used to classify vehicle collision data.

The inventive approach allows a meaningful application of a machine-readable method in the detection of vehicle side collisions. Viewed in general terms, the approach of the invention enables the meaningful application of a machine-readable method in algorithms consisting of independent, equal partial algorithms.

The machine-based learning method can be characterized by an automatic calculation of characteristics for the separation of different classes of vehicle collisions and by the applicability of the machine-based learning method not only as 2-dimensional but also to 3-dimensional representations.

The learning method requires as an input channel not only suitably processed sensor signals, but also a control channel. The control channel has the property of being a monotonically increasing function via the collision course. For example, the control channel may be based on an integral calculated from the ECU-X sensor signal, i. from a signal of an acceleration sensor of the airbag control unit oriented in the vehicle front-rear direction. The sensor providing the ECU-X sensor signal may be the most important sensor of a frontal algorithm, that is, an algorithm for classifying a frontal collision. Other channels, such as ECU-Y, UFS1 and UFS2 can be used only as input for the classification. The channels UFS1 and UFS2 can each refer to Umeldsensoren.

By means of the control channel, a so-called evaluation range can be defined, which specifies a time range for the separation of the relevant collision classes. Both the training of the machine-based learning method (offline training) and the classification of the sensor signals (online on ECU) can only take place within this control channel determined by the Start & Stop area.

1. (a) über den gesamten Kollisionsverlauf monoton ansteigend ist,
2. (b) das er für eine große Anzahl unterschiedlicher Kollisionstypen verfügbar ist und
3. (c) das er innerhalb der geforderten Auslösezeit ausreichend hohe Werte erreicht.

In the course of applying the machine-based learning method not only to frontal collisions but also to page collision data, an appropriate control channel is required. This should have the property that he

1. (a) is monotonically increasing over the entire collision course,
2. (b) it is available for a large number of different collision types, and
3. (c) that it reaches sufficiently high values within the required triggering time.

• (d) das linksseitige und rechtsseitige Kollisionen vom selben Typ in einen vergleichbaren Signalverlauf des Steuerkanals führen.

In addition, with regard to side collision detection, there is an additional requirement

• (d) the left-side and right-side collisions of the same type lead to a comparable signal course of the control channel.

The requirements a) - c) can be adopted by the integrated ECU-X sensor signal for use in the frontal collision classification. The sensor providing the ECU-X sensor signal may at the same time be the only authorized triggering sensor. The triggering decision can only be supported by further sensors, for example ECU-Y, UFS1, UFS2.

However, in the side-collision algorithm, a large number of sensors and identical partial algorithms based thereon are involved equally in a triggering decision. This is caused by very short trip times, since only the sensor located near the point of impact causes a relevant signal within the required time, as well as caused by different points of impact for collisions and the associated different necessary sensor positions. The sensor signals measured in this case fulfill the requirement a) even after additional processing only in a few cases, but are rather oscillating or have too low a signal amplitude.

Furthermore, not only the various sensor positions play a role, such as PAS arranged in the B-pillar, PAS arranged in the C-pillar or ECU-Y in the central control unit, but also the sensor type, so for example, whether it is an acceleration sensor or a pressure sensor acts. PAS denotes a paged acceleration sensor that provides the lateral acceleration of the vehicle. Experience has shown that none of the sensor positions or sensor types mentioned alone is capable of fulfilling all requirements a) -d).

The present invention solves the problem of requirements a) -d) by a suitable combination of all sensor signals.

The present invention provides a method of combining sensor data for collision classification of a vehicle comprising the steps of: receiving temporal traces of sensor signals via an interface, the sensor signals representing signals provided by different sensors; and combining the sensor signals to obtain a time history of a combined signal for classifying the collision

The collision can be a frontal collision or a side collision. The side collision can be characterized by the action of a force acting on the vehicle in the vehicle transverse direction. The classification can be used to determine what type of collision it is. For example, it can be determined at which position the force is applied to the vehicle and how strong the force is. Depending on the classification, vehicle occupant protection means adapted to the collision can be triggered. The inventive method can be implemented in a control unit. Accordingly, the interface may be an interface of the controller. If all or some of the sensors are arranged in the control unit, the interface may be an internal interface. The different sensors can be sensors of the same type, which are arranged at different positions in the vehicle or have different detection ranges. Also, the different sensors may include different sensor types, such as pressure sensors, structure-borne sound sensors or acceleration sensors, which are designed to detect different physical quantities. By the different sensors can be detected both in the longitudinal direction and in the transverse direction acting on the vehicle forces. The time profile of a sensor signal can represent a change over time of a measured variable which is detected by a respective sensor. A resolution of the sensor signals can be adapted to the respective measured variable. All or at least some of the sensor signals have a resolution that can represent more than two states. This means for a digitally present sensor signal that it has a resolution of at least 2 bits. Combining the sensor signals may include adding the data provided by the individual sensors. For this purpose, the sensor signals can be supplied to a suitable combination device or addition device. In each case, data can be combined which are detected or provided by the sensors at the same time. Before combining or during combining, the sensor signals can be matched to each other. For example, the sensor signals can be scaled or normalized. By combining, from the plurality of individual sensor signals, a single combined sensor signal may be determined that includes the information of the individual sensor signals. For combining, the original, the scaled or the normalized time profiles of the sensor signals can be combined with one another. In particular, the time profiles can be combined without a logical combination of the sensor signals being carried out. The temporal courses can have a continuous or a discrete course of values. If the time courses have a continuous course of values, as may be the case, for example, with an analog signal, the combination can also be carried out continuously. If the time courses have a discrete course of values, as may be the case, for example, with a digital signal, then discrete values of the time courses associated with a respective time can be combined with one another. The combined sensor signal may over time have an alternating rising and falling course. For example, the combined sensor signal can oscillate. Alternatively, the combination may be performed so that the combined sensor signal has a monotone character. The combined signal may be provided to a classification device or to an evaluation device in which a classification of the collision based on the combined signal may be performed. The classification can be carried out by means of a machine-based learning method. The machine-based learning method may include a support vector machine. Thus, the combined signal may represent an input signal or control channel for the machine-based learning method.

According to one embodiment, in the step of combining, a monotonization of the time profiles of the sensor signals or of the time profile of the combined signal can be carried out. By monotonizing a monotonized combined signal can be determined, which has a monotonously increasing over time signal waveform or has a monotonically falling over time waveform. To determine the monotonized combined signal, first the sensor signals can be combined to obtain a combined signal which can then be monotonized. Alternatively, all or part of the sensor inputs may be monotonized prior to combining.

According to one embodiment, the method according to the invention may comprise a step of evaluating the combined signal by means of a machine-based learning method in order to be able to carry out the classification of the collision. The use of a machine-based learning method is advantageous because it allows automatic calculation of characteristics for separating different classes of vehicle collisions. In addition, the machine-based learning method is also applicable to three-dimensional representations.

Furthermore, the method may comprise a step of scaling the sensor signals to a common value range. This allows an adaptation of the data provided by the sensor signals, for example with regard to their range of values or their resolution. In this way, sensor signals provided by different sensors can be related to each other with respect to the collision.

According to one embodiment, a first set of the sensor signals may be associated with a first physical quantity and a second set of the sensor signals may be associated with a second physical quantity. In a step of normalization, at least one of the sets of sensor signals can be normalized to one of the first set and the second set of common value bases. Sensors configured to provide the first set of sensor signals may be of a different type or based on a different measurement method than the sensors configured to provide the second set of sensor signals. For example, the first set of sensor signals from pressure sensors and the second set of sensor signals from acceleration sensors may be provided. The pressure changes and accelerations occurring in one and the same collision can have a different time course and different amplitudes. Also, the time course depending on the severity of Collision are different. By means of standardization, the sensor signals of the first quantity and the second quantity can be adapted to one another in such a way that a combination of the sensor signals that makes sense with respect to the collision is possible. The standardization may include, for example, a gain of individual sensor signals. The normalization can be done by means of a time-dependent normalization function. In this way, a standardization with regard to the severity of the collision or with regard to the measuring methods used can be carried out.

According to one embodiment, a first of the sensor signals may represent a signal provided by a first sensor located at a first position in the vehicle, and a second of the sensor signals may represent a signal provided by a second sensor disposed at a second position in the vehicle , In a step of normalization, the first sensor signal and / or the second sensor signal can be normalized based on a distance between the first sensor and the second sensor. For example, the first sensor may be arranged peripherally in the vehicle and the second sensor centrally in the vehicle. Also, the first sensor and the second sensor may be arranged on opposite sides of the vehicle. For example, one of the sensors may be arranged on the right side of the vehicle and one of the sensors on the left side of the vehicle. In this way, the fact that the collision on the different sides of the vehicle as well as peripherally and centrally in each case has different effects and thus leads to different temporal courses of the sensor signals can be taken into account.

According to one embodiment, the method according to the invention may comprise a step of activating a time window. The activation of the time window can take place if at least one of the sensor signals fulfills a predetermined start condition. The combining of the sensor signals may be carried out when the time window is activated. The time window can be activated when an occurrence of a collision is detected. The fulfillment of the start condition can be detected by means of a threshold value decision. Accordingly, the predetermined starting condition may correspond to a threshold value with which the at least one of the sensor signals is compared. If the threshold value is exceeded, it can be assumed that there is a collision. For each of the sensor signals, which are evaluated to activate the time window, a separate predetermined starting condition can be defined. By activating the time window, a time channel can be started which can be used to classify the collision. The time channel may correspond to a time signal in which the time runs from a start value at the beginning of the time window.

Furthermore, the method according to the invention may comprise a step of evaluating at least one of the sensor signals in order to assign a vehicle side to the collision. In a step of providing information about the vehicle side can be provided. The information about the vehicle side may be provided in addition to the combined signal. The information about the vehicle side may be used to classify a side collision because a characteristic of the combined signal may be independent of the collision side. The at least one sensor signal may be provided by an acceleration sensor or by a pressure sensor. If a plurality of sensor signals are evaluated, these may include both acceleration sensor signals and pressure sensor signals.

The present invention further provides an apparatus for combining sensor data for classifying a collision of a vehicle, comprising: means for receiving temporal traces of sensor signals via an interface, the sensor signals representing signals provided by different sensors; and means for combining the timings of the sensor signals to obtain a time history of a combined signal for classifying the collision.

The device according to the invention can be designed to carry out or implement the steps of the method according to the invention in corresponding devices. Also by this embodiment of the invention in the form of a device, the object underlying the invention can be solved quickly and efficiently.

In the present case, a device can be understood as meaning an electrical device which processes sensor signals and outputs control signals in dependence thereon. The device may have an interface, which may be formed in hardware and / or software. In the case of a hardware-based embodiment, the interfaces can be part of a so-called system ASIC, for example, which contains a wide variety of functions of the device. However, it is also possible that the interfaces are their own integrated circuits or at least partially consist of discrete components. In a software training For example, the interfaces can be software modules that are present, for example, on a microcontroller in addition to other software modules.

Also of advantage is a computer program product with program code which is stored on a machine-readable carrier such as a semiconductor memory, a hard disk memory or an optical memory and is used to carry out the method according to one of the embodiments described above, when the program is executed on a control unit.

• 1 ein Blockschaltbild eines Ausführungsbeispiels der vorliegenden Erfindung;
• 2 ein Ablaufdiagramm eines Ausführungsbeispiels der vorliegenden Erfindung;
• 3 eine Anordnung von Sensoren in einem Fahrzeug, gemäß einem Ausführungsbeispiel der vorliegenden Erfindung;
• 4 eine Anordnung von Sensoren in einem Fahrzeug, gemäß weiteren Ausführungsbeispielen der vorliegenden Erfindung;
• 5 eine Darstellung von Signalverläufen, gemäß einem Ausführungsbeispiel der vorliegenden Erfindung;
• 6 eine Darstellung von weiteren Signalverläufen, gemäß einem Ausführungsbeispiel der vorliegenden Erfindung; und
• 7 eine Darstellung von weiteren Signalverläufen, gemäß einem Ausführungsbeispiel der vorliegenden Erfindung.

The invention will now be described by way of example with reference to the accompanying drawings. Show it:

• 1 a block diagram of an embodiment of the present invention;
• 2 a flowchart of an embodiment of the present invention;
• 3 an arrangement of sensors in a vehicle, according to an embodiment of the present invention;
• 4 an arrangement of sensors in a vehicle, according to further embodiments of the present invention;
• 5 a representation of signal waveforms, according to an embodiment of the present invention;
• 6 a representation of further signal waveforms, according to an embodiment of the present invention; and
• 7 a representation of further signal waveforms, according to an embodiment of the present invention.

In the following description of preferred embodiments of the present invention, the same or similar reference numerals are used for the elements shown in the various figures and similarly acting, wherein a repeated description of these elements is omitted.

1 shows a block diagram of an implementation of a method for combining sensor data for a classification of a collision, for example a side collision, of a vehicle 100 , according to an embodiment of the present invention. In the vehicle 100 Sensors are arranged, of which, for example, a first sensor 102 , a second sensor 104 and a third sensor 106 are shown. The sensors 102 . 104 . 106 are designed to detect physical quantities that occur in a collision of the vehicle. For example, the sensors 102 . 104 . 106 Accelerations of the vehicle in an x-direction, so a vehicle longitudinal direction, and in a y-direction, so detect a vehicle transverse direction. Furthermore, the sensors 102 . 104 . 106 be formed to detect structure-borne noise, which occurs in a collision of the vehicle. The sensors 102 . 104 . 106 may be configured to provide sensor signals. The sensor signals may be from a combination device 110 , via an appropriate interface, are received. The combination device 110 can be part of a control unit of the vehicle 100 be. The combination device 110 may be configured to combine the received sensor signals with each other, and thus to produce a combined signal. Furthermore, the combination device 110 be formed, and to monotize the combined signal and thus provide a monotonized signal having, for example, starting from a time of detecting a collision, a monotonously increasing waveform. The monotonized signal may be sent to a classifier 115 to be provided. The classifier 115 may be configured to classify the collision based on the monotonized signal. For this purpose, the classification device 115 be designed to implement a machine-based learning process. The monotone signal may represent a control channel or part of a control channel for the machine-based learning method. Alternatively, the combined signal can also be used as the control channel for the machine-based learning method. According to the classification of the collision, the collision can be correspondingly triggered in the occupant protection means, of which, for example, a first occupant protection means 122 and a second occupant protection means 124 are shown. At the occupant protection devices 122 . 124 For example, these may be belt tensioners or airbags.

2 FIG. 12 is a flow chart of a method of combining sensor data for classifying a collision of a vehicle according to an embodiment of the present invention. FIG. According to the method, in one step 201 Sensor signals are received via an interface. The sensor signals may be signals provided by different sensors become. For example, the sensor signals from the in 1 be provided sensors shown. The sensor signals can be received continuously. In one step 203 the sensor signals are combined to obtain a combined signal. In this case, all received sensor signals or a selection of the received sensor signals can be combined. The combining can be carried out continuously. For example, the combining may be performed in response to a start signal. The start signal may be provided in response to a detected collision of the vehicle. In one step 205 the combined signal is combined to obtain a monotone signal. The steps of combining 203 and monotonizing 205 can also be performed in a common step to generate from the sensor signals the monotonized signal. Based on the monotonized signal, the collision can be classified.

3 shows a vehicle 100 with an array of sensors 102 . 104 . 106 , whose sensor signals according to an embodiment of the present invention can be used to detect and classify a collision of the vehicle according to the invention. At the sensor 102 it can be a first environment sensor (UFS) and the sensor 106 it can be a second environment sensor. The sensors 102 . 106 may be configured to detect a size acting in the vehicle longitudinal direction. At the sensor 104 can be a control unit (ECU), in which a corresponding sensor is arranged. For example, the control unit may be an airbag control unit. The sensor 104 can be designed to detect both a vehicle-longitudinal direction acting in the vehicle longitudinal direction and size. If the sensor 104 is designed as a control unit, the steps of the method can be partially or completely implemented on the control unit.

4 shows a vehicle 100 with another arrangement of sensors 104 , 402, 406, 407, 408, whose sensor signals according to an embodiment of the present invention can be used to detect and classify a collision of the vehicle according to the invention. The sensors 402 , 407 may be arranged on the right side of the vehicle, with the sensor 402 in the front part of the vehicle and the sensor 407 can be arranged in the rear part of the vehicle. The sensors 406 . 408 can be arranged on the left side of the vehicle, the sensor 406 in the front part of the vehicle and the sensor 408 may be arranged in the rear part of the vehicle. For example, the sensors 402 . 406 in the respective B-pillar and the sensors 407 . 408 be arranged in the respective C-pillar of the vehicle. At the sensor 104 can become appropriate too 3 to act a controller in which a suitable sensor is arranged.

According to one embodiment, the sensor may be 402 around a first acceleration sensor PAS1, at the sensor 406 around a second acceleration sensor PAS2, at the sensor 407 around a third acceleration sensor PAS3 and at the sensor 408 to act a fourth acceleration sensor PAS4. The sensors 402 . 406 . 407 . 408 can each detect a lateral acceleration of the vehicle. In this case, the sensors 402, 407 and the sensors 406 . 408 evaluate the lateral accelerations with opposite signs.

According to a further embodiment, it may be in the sensor 402 around a first pressure sensor PPS1, at the sensor 406 around a second pressure sensor PPS2, at the sensor 407 around the third acceleration sensor PAS3 and the sensor 408 to act on the fourth acceleration sensor PAS4. The sensors 407 . 408 can each detect a lateral acceleration of the vehicle. The sensors 407 . 408 For example, each may be disposed in a vehicle door and configured to meter the air pressure in the door. For example, a dynamic pressure change may be detected by a door deformation in the side collision and additionally an absolute pressure.

Thus, based on 4 exemplary sensor configurations shown, with other sensor arrangements and other sensor configurations are also possible.

According to a first sensor configuration, four times acceleration sensors are exemplified 402 . 406 . 407 . 408 PAS and once a lateral acceleration sensor ECU-Y of the ECU sensor 104 used. On the sensors 104 , 402, 406, 407, 408 equal and similar sub-algorithms can be calculated. By means of the partial algorithms, a pre-processing of the sensor signals can already be carried out.

According to a second sensor configuration, acceleration sensors are twice 407 . 408 PAS and two pressure sensors 402 . 406 PPS and once a lateral acceleration sensor ECU-Y of the ECU sensor 104 used.

According to a third sensor configuration, not shown, only one pressure sensor PPS and one lateral acceleration sensor ECU-Y of the control unit sensor can also be used 104 to be available.

None of the described sensor positions alone is capable of fulfilling all requirements a) -d).

The inventive approach solves the problem of the requirements a) - d) by a suitable combination of all sensor signals.

The approach according to the invention comprises several variants of a control channel, which can be used for example by a machine-based learning method for classifying a collision.

A first variant relates to a global time channel. A second variant relates to a normalized addition of all existing sensor signals. As an additional function, a distinction between a left-side and a right-side collision can be imaged with. This can be realized by means of additional thresholds.

5 shows a representation of three diagrams, with reference to which an activation of the time channel is described according to an embodiment of the present invention.

In a first diagram is plotted on the ordinate, the time t and on the abscissa a range of values of a sensor signal ECU-Y 531, for example, of the in 4 shown sensor 104 can be provided. Further, a sensor signal 531 assigned threshold 533 (Start Thd) shown.

In a second diagram is on the ordinate, the time t and on the abscissa, a range of values of a sensor signal PAS 535 applied, for example, by the in 4 shown sensor 402 can be provided. Further, a sensor signal 535 assigned threshold 537 (Start Thd) shown.

In a third diagram, time is t on the ordinate and value range of the global time channel on the abscissa 540 (global timer).

The sensor signal 531 , which may optionally be preprocessed, has a wave-shaped course and exceeds at a first time 542 the threshold 533 , By crossing the threshold 533 a time window is activated and the global time channel 540 started. The threshold 533 and in a similar way the threshold 537 , thus act as start thresholds for the global time channel 540 , At a second time 544. falls below the sensor signal 531 the threshold 533 , Falling below the threshold 533 causes the beginning of a holding time 546 , A duration of the holding time 546 can for each of the sensor signals 531 . 535 be equal or different. According to this embodiment, the hold time is running 546 at a third time 548. The expiration of the holding time 546 causes a deactivation of the time window and thus of the global time channel 540 , The global time channel 540 can be formed by a monotonously increasing time signal. Thus, a value of the global time channel may be 540 starting from, for example, a zero output value from the first time 542 continuously increase until the value of the global time channel 540 at the third time 548 back to baseline. The value of the global time channel 540 can then remain at the output value until the sensor signal 531 again the threshold 533 or alternatively until the sensor signal 535 the threshold 537 exceeds. As an alternative to the threshold decision, other suitable decision bases can be used to start the global time channel 540 be used. The decision base can be chosen in each case so that the global time channel 540 starts when the occurrence of a collision event can be assumed.

With the start of the global time channel 540 can be started with the combination of the sensor signals. Alternatively, the combination of the sensor signals and the monotonization of the combined sensor signal can be carried out continuously, wherein an evaluation of the monotonized signal can be started with the start of the global time channel. The time channel 540 can be used as a time scale for the combined signal. Furthermore, the time channel 540 in addition to the combined signal to be used together with the combined signal as the basis for classifying the collision.

According to the in 5 In the embodiment shown, the global time channel 540 starts when at least one sensor signal 531 . 535 a sensor-dependent threshold 533 . 537 has exceeded. The global time channel 540 is incremented until none of the sensor signals 531 . 535 more the respective ones sensor-dependent threshold 533 . 537 exceeds and either a central or sensor dependent activation hold time 546 has expired. Thereafter, the time channel 540 is reset to zero. The time channel 540 can also be set to zero if multiple collision is detected, ie, by way of example, a right-sided collision directly follows a left-side collision.

6 shows a representation of four diagrams, with reference to which a combination of sensor signals and a monotonization according to an embodiment of the present invention will be described.

In a first diagram is plotted on the ordinate time t and on the abscissa a range of values of a sensor signal ECU-Y 531, which corresponds to the in 5 can correspond to the signal shown.

In a second diagram is on the ordinate, the time t and on the abscissa, a range of values of a sensor signal PAS 535 applied in the 5 can correspond to the signal shown.

In a third diagram, time is t on the ordinate and value range of a combined signal on the abscissa 652 applied. The combined signal 652 corresponds to a combination of the signals 531 . 535 , For example, the combined signal 652 by adding the signals 531 . 535 be determined. Due to the wavy course of the signals 531 . 535 has the combined signal 652 also a wave-shaped course. The signals 531, 535 are chosen by way of example. According to the inventive approach further sensor signals can be taken into account, which then in addition to the signals 531 . 535 can be combined to the combined signal 652 to determine. Before or during the combination, the signals 531 . 535 scaled and normalized. That way, the signals can 531 . 535 both in terms of their value range and in terms of their timing, so that the information of each one of the signals 531, 535, with respect to the rest of the signals 531 . 535 , in the right proportion in the combined signal 652 with flows.

In a fourth diagram, the ordinate is the time t and the abscissa a value range of a combined and monotonized signal 654 applied. The monotonized signal 654 can by monotonizing the combined signal 652 from the combined signal 652 be determined. For example, the combined signal 652 integrated or low-pass filtered to the monotonized signal 654 to create. Also, maximum values of the combined signal can occur 652 each held until a subsequent larger maximum value occurs.

$$\text{Signal\_Combined\_2}=\text{f}\left(\text{a}*\text{g}\left(\text{PAS}\left(\text{t}\right)\right)+\text{b*h}\left(\text{ECU}-\text{Y}\left(\text{t}\right)\right)\right)$$

According to one embodiment, the combined signal 652 by means of a normalized addition of the signals 531 . 535 be determined. The normalized addition is a construction for combining different sensors of one type and different sensor types. The signals of a type values are normalized with respect to their resolution (eg ECU-Y = 5 LSB / g, PAS = 4 LSB / g). After successful combination, the signal must still be suitably monotonized before use as a control channel. For monotonizing, an integral can be formed, a strong low pass can be used or a temporary holding of maximum sensor values can be carried out. This can, as with the help of 6 represented, according to the following general formulas: $$\text{Signal\_Combined\_1}=\text{a}*\text{f}\left(\text{PAS}\left(\text{t}\right)\right)+\text{b}*\text{f}\left(\text{ECU}-\text{Y}\left(\text{t}\right)\right)$$

$$\text{Signal\_Combined\_2}=\text{f}\left(\text{a}*\text{G}\left(\text{PAS}\left(\text{t}\right)\right)+\text{bra}\left(\text{ECU}-\text{Y}\left(\text{t}\right)\right)\right)$$

In this case, f (t), g (t), h (t) are functions for monotonizing and the coefficients a and b are used to adapt the sensor resolution. The Signals_Combined_1 and Signal_Combined_2 signals can be examples of the in 6 signal shown 654 represent.

In order to determine the signal Signal_Combined_1, the time profile of the sensor signal PAS (t) can therefore be monotonized by means of the monotonizing function f (t) and multiplied by the coefficient a in order to obtain a normalized and monotonized profile of the sensor signal PAS (t). Furthermore, the time profile of the sensor signal ECU-Y (t) can be monotonized by means of the monotonizing function f (t) and multiplied by the coefficient b to obtain a normalized and monotonized curve of the sensor signal ECU-Y (t). Subsequently, the respective normalized and monotonized curves of the sensor signals PAS (t) and ECU-Y (t) can be added together.

In order to determine the signal Signal_Combined_2, therefore, the time profile of the sensor signal PAS (t) can be monotonized by means of the monotonizing function g (t) and multiplied by the coefficient a to obtain a normalized and monotonized profile of the sensor signal PAS (t). Furthermore, the time profile of the sensor signal ECU-Y (t) can be monotonized by means of the monotonizing function h (t) and multiplied by the coefficient b to obtain a normalized and monotonized curve of the sensor signal ECU-Y (t). Subsequently, a sum of the respective normalized and monotonized profiles of the sensor signals PAS (t) and ECU-Y (t) can be formed and then the sum can be monotonized with the monotonizing function f (t).

$$\text{Signal\_Combined\_1b}=\text{f}\left(\text{a}*\text{PAS}\left(\text{t}\right)+\text{b}*\text{ECU}-\text{Y}\left(\text{t}\right)+\text{y}\left(\text{PPS}\left(\text{t}\right)\right)\right)$$

$$\text{Signal\_Combined\_2}=\text{f}\left(\text{g}\left(\text{a}*\text{PAS}\left(\text{t}\right)\right)+\text{h}\left(\text{b}*\text{ECU}-\text{Y}\left(\text{t}\right)\right)+\text{k}\left(\text{y}\left(\text{PPS}\left(\text{t}\right)\right)\right)\right)$$

If different sensor types are used, the value normalization with regard to the collision severity should be done. The goal of normalization is the generation of a comparable amplitude of the sensor types used, that is, for example, an acceleration sensor and a pressure sensor, for the same collision severity. This can be implemented according to the following formulas: $$\text{Signal\_Combined\_1a}=\text{a}*\text{f}\left(\text{PAS}\left(\text{t}\right)\right)+\text{b}*\text{G}\left(\text{ECU}-\text{Y}\left(\text{t}\right)\right)+\text{k}\left(\text{y}\left(\text{PPS}\left(\text{t}\right)\right)\right)$$

$$\text{Signal\_Combined\_1b}=\text{f}\left(\text{a}*\text{PAS}\left(\text{t}\right)+\text{b}*\text{ECU}-\text{Y}\left(\text{t}\right)+\text{y}\left(\text{PPS}\left(\text{t}\right)\right)\right)$$

$$\text{Signal\_Combined\_2}=\text{f}\left(\text{G}\left(\text{a}*\text{PAS}\left(\text{t}\right)\right)+\text{H}\left(\text{b}*\text{ECU}-\text{Y}\left(\text{t}\right)\right)+\text{k}\left(\text{y}\left(\text{PPS}\left(\text{t}\right)\right)\right)\right)$$

In this case, f (t), g (t), h (t), k (t) are functions for monotonization, the coefficients a and b are used for adaptation of the sensor resolution and y (t) is a normalization function between acceleration and pressure sensors. The signals Signal_Combined_1a, Signal_Combined_1b and Signal_Combined_2 can be examples of the in 6 signal shown 654 represent.

In order to determine the signal Signal_Combined_1 a, the time profile of the sensor signal PAS (t) can therefore be monotonized by means of the monotonizing function f (t) and multiplied by the coefficient a in order to obtain a normalized and monotonized profile of the sensor signal PAS (t). Furthermore, the time profile of the sensor signal ECU-Y (t) can be monotonized by means of the monotonizing function g (t) and multiplied by the coefficient b to obtain a normalized and monotonized curve of the sensor signal ECU-Y (t). Furthermore, the time profile of the sensor signal PPS (t) can be normalized by means of the normalization function y (t) in order to obtain a normalized profile of the sensor signal PPS (t). Subsequently, the normalized sensor signal y (PPS (t)) can be monotonized with the monotonizing function k (t). Subsequently, a sum of the respectively monotonized progressions of the sensor signals can be formed.

In order to determine the Signal_Combined_1b signal, the time profile of the sensor signal PAS (t) can therefore be multiplied by the coefficient a in order to obtain a normalized profile of the sensor signal PAS (t). Further, the timing of the sensor signal ECU-Y (t) may be multiplied by the coefficient b to obtain a normalized history of the sensor signal ECU-Y (t). Furthermore, the time profile of the sensor signal PPS (t) can be normalized by means of the normalization function y (t) in order to obtain a normalized profile of the sensor signal PPS (t). Subsequently, the normalized curves of the sensor signals PAS (t), ECU-Y (t), PPS (t) can be added and then monotonized with the monotonizing function f (t).

7 shows a representation of three diagrams, on the basis of which a collision side, according to an embodiment of the present invention, is determined.

In a first diagram the time t is on the ordinate and a value range of a signal on the abscissa 761 a machine-based learning (SVM) process that indicates whether a trip collision is detected. A deployment collision is a collision in which occupant protection devices are triggered. According to this embodiment, the signal 761 an increase associated with the detection of a trip collision.

In a second diagram, the time is t on the ordinate and a value range of a signal on the abscissa 763 plotted indicating whether a left-side collision is detected. The signal 763 stays at zero. Thus, no left-side collision was detected.

In a third diagram, the time is t on the ordinate and a value range of a signal on the abscissa 765 which indicates whether a right-hand collision is detected. The signal 765 has an increase associated with the detection of a right-hand collision. Thus, a right-sided collision was detected and responsive to the signal change of the signal 761 the occupant protection means assigned to a right-side collision can be triggered.

To meet requirement d), both the machine-readable method and the control channel must be independent of the collision side. This means that with the same type of collision but different collision side both the incoming signals and the control channel must be comparable. This results in a possible normalization on the collision side of both the control channel and the other signal characteristics used for classification.

Furthermore, the final decision is missing when taken trigger decision, whether left or right to be triggered. This decision must therefore additionally be made by further comparisons between sensor signals and left-side collision thresholds or right-side collision thresholds, as described in US Pat 7 is shown.

The part of the inventive approach relating to requirement d) is optional. It depends, among other things, on whether only one or more 3-dimensional classifiers are available.

The embodiments described and shown in the figures are chosen only by way of example. Different embodiments may be combined together or in relation to individual features. Also, an embodiment can be supplemented by features of another embodiment. Furthermore, method steps according to the invention can be repeated as well as carried out in a sequence other than that described.

Claims (9)

A method for combining sensor data for a classification of a collision of a vehicle (100), comprising the following steps: receiving (201) time profiles of sensor signals (531, 535) via an interface, wherein the sensor signals from different sensors (102, 104, 106, 402, 406, 407, 408) represent signals provided; and combining (203) the time histories of the sensor signals to obtain a time history of a combined signal (654) for classifying the collision, characterized in that a first set of sensor signals (531, 535) of a first physical quantity and a second quantity associated with the sensor signals of a second physical quantity, and with a step of normalizing at least one of the sets of sensor signals to match the sensor signals of the first set and the second set prior to the step of combining (203). Method according to Claim 1 in which, in the step of combining, monotonization (205) of the time profiles of the sensor signals (531, 535) or of the time profile of the combined signal (654) is carried out. Method according to one of the preceding claims, comprising a step of evaluating the monotonized signal (654) by means of a machine-based learning method to perform the classification of the collision. Method according to one of the preceding claims, comprising a step of scaling the sensor signals (531, 535) to a common value range. Method according to one of the preceding claims, in which a first of the sensor signals represents a signal which is provided by a first sensor (102, 106; 402, 406, 407, 408) arranged at a first position in the vehicle and a second one of the sensor signals Signal provided by a second sensor (104) disposed at a second position in the vehicle and including a step of normalizing the first sensor signal and / or the second sensor signal based on a distance between the first sensor and the second sensor. The method of any one of the preceding claims, including the step of activating a time window (540) when at least one of the sensor signals meets a predetermined start condition and wherein the combining of the sensor signals is performed when the time window is activated. Method according to one of the preceding claims, comprising a step of evaluating at least one of the sensor signals (531, 535) to associate the collision with a vehicle side and a step of providing information about the vehicle side, in addition to the combined signal (654). Device for combining sensor data for a classification of a collision of a vehicle (100), having the following features: a device for receiving (201) time profiles of sensor signals (531, 535) via an interface, wherein the sensor signals from different sensors (102, 104 , 106, 402, 406, 407, 408) represent signals provided; and means (203) for combining the temporal characteristics of the sensor signals to obtain a time characteristic of a combined signal (654) for classifying the collision, characterized in that a first quantity of the sensor signals (531, 535) of a first physical quantity and a second set of the sensor signals are associated with a second physical quantity and the means for combining (203) is arranged to normalize at least one of the sets of the sensor signals to match the sensor signals of the first set and the second set prior to combining. Computer program product with program code, which is stored on a machine-readable carrier, for carrying out the method according to one of Claims 1 to 7 when the program is running on a controller.

Images