DE102015005961A1 - Method for monitoring the measurement signals of at least one sensor - Google Patents

Abstract

The invention relates to a method for monitoring the measuring signal of a sensor (2, 3, 4) from a sensor network having at least two sensors (2, 3, 4) which have an at least partially overlapping measuring space, and which have the same target object with different physical measuring principles (5) record. The inventive method is characterized in that a cross comparison of the measurement signals of the individual sensors (2, 3, 4) of the sensor network is carried out, wherein the mean values of the deviations of the measurement signals are detected to each other, and wherein in a permanent deviation of the average values from zero to a relative decalibration of one of the sensors (2, 3, 4) is closed. The preferred use is in the field of environment detection for autonomous driving.

Description

The invention relates to a method for monitoring the measurement signals of at least one sensor according to the type defined in more detail in the preamble of claim 1. The invention also relates to the use of such a method according to claim 8.

The safe function of sensors is indispensable for a variety of applications. Most sensors used have diagnostic features for self-monitoring. This can reliably exclude some of the potential sources of error within the sensor. In addition, sensors can also be present several times in order to achieve a higher level of security through this redundancy.

It is now the case in practice that measurement signals can also be generated by a sensor which occur, for example, as a result of a misalignment of the sensor, in particular due to, for example, an impairment of its mechanical suspension. In this case, the self-diagnosis of the sensor can not detect an error, potentially operating on the basis of erroneous measurement signals. For example, in the case of safety-critical applications, in particular of applications in the field of autonomous driving of logistics vehicles, commercial vehicles or passenger cars, this can represent a considerable risk.

In the publication M. Gabb, S. Krebs, O. Löhlein and M. Fritzsche, "Probabilistic Interference of Visibility Conditions by Means of Sensor Fusion", in IEEE Intelligent Vehicles Symposium, 2014 Therefore, it is described how the visibility of a monocular camera can be verified by a radar sensor. The visual problems of the camera with corresponding weather conditions such as fog, heavy rain or snowfall do not affect the radar sensor as the second sensor, so that a comparison of the data of these sensors, a monitoring of the camera signals and a determination of the plausibility of the corresponding measured values is possible. The problem here is the fact that, although the weaknesses of the individual sensor types are known per se. However, an unexpected deviation of the one or the other sensor value does not permit any statement as to which sensor supplies the faulty or impaired measuring signals without exact knowledge of the influence.

From the US 2014/0307247 A1 is the signal processing of a composite of sensors known, which can also be used for environment detection of a vehicle, in particular for autonomous driving. The measurement signals of the individual sensors are processed with each other and compared with each other to infer due to the different physical measurement principles of the sensors on certain weather conditions, such as moisture in the environment of the vehicle. In particular, the signal of a laser scanner or LIDAR can be adjusted with the signal of a radar and / or a camera and, if necessary, other sensors, so as to appropriate weather conditions, such as a wet road and the associated problems in terms of tire adhesion, the visibility restriction etc., to be able to conclude.

For further general state of the art is also to the US patent application US 2014/0333468 A1 pointed.

Now it is the case, in particular for the application in the autonomous driving of vehicles, that the highest safety requirements apply to the sensors and a high reliability of the detected sensor data has to be ensured, since otherwise this can very quickly be associated with a considerable risk with regard to the safety of the autonomously driven vehicle ,

For this reason, it is already customary today for the sensors in such systems to be of correspondingly high quality, and for these to be used in such a way that self-diagnosis of the sensors is carried out, in particular when starting the system. About such a self-diagnosis of the sensors, in particular the reliable functionality of the sensor itself, as well as the functioning of the detection devices used of the sensor can be verified. However, it is a serious problem that a shift of parts of the sensors with each other or the sensors themselves against each other by such a self-diagnosis of the sensor can not always be completely detected and thus corrected. If, for example, a camera sensor is correspondingly misadjusted in relation to a radar sensor and / or a laser scanner or LIDAR, it can in the worst case happen that the same target object can no longer be recognized as the same target object during the detection of the sensor signals, but that different target objects are assumed. In case of doubt, this can lead to safety-critical situations, in particular when used for autonomous driving systems. It is also the case that such a misalignment of the sensors by a mechanical displacement of the position of the sensors can also occur during operation, in particular by temperature effects and the associated different thermal Expansions. An impairment due to releasing fastenings of the sensors and / or a deformation, for example due to a collision or the like, are conceivable as possible causes. None of the known methods can reliably and reliably detect such an impairment, in particular if it occurs during the operation of a vehicle equipped with the sensors.

It is therefore an object of the present invention to provide a method which is able to detect such a decalibration of a sensor from a composite of at least two sensors safely and reliably, so that appropriate measures can be taken.

According to the invention this object is achieved by a method having the features in the characterizing part of claim 1. Advantageous embodiments and further developments emerge from the subclaims dependent thereon. In addition, a particularly preferred use of the method is specified in claim 8. Again, advantageous embodiments and developments of the dependent claims.

The method according to the invention uses the measurement signals from at least two sensors of a sensor network in order to detect the mean values of the deviation of the measurement signals from one another via a cross comparison. With a permanent deviation of the mean values from zero, it is then possible to conclude that a relative decalibration of one of the sensors is involved. By means of this method, a decalibration of one of the sensors, for example a squinting stereo camera which is mutually misaligned with respect to its optical axes, can thus be detected reliably and reliably, whereas such an error could not be detected during the self-diagnosis of the sensor. As a result, the decalibration of a sensor can be detected quickly and reliably during operation. This includes rapid changes as well as drift behavior over longer periods, which can be reliably detected. Any necessary countermeasures, such as an emergency shutdown of the operation or the like, can be taken.

According to a very advantageous development of the idea, it is further provided that the sensor network comprises at least three sensors with different physical measurement principles. In such a particularly advantageous construction of the sensor network comprising at least three sensors with different physical measurement principles, in addition to the basic recognition of a decalibration present in the sensor network, it can also be deduced which of the sensors has been decalibrated so that the error occurring in the sensor data group is concretely one of the errors Assign sensors.

According to a very simple and advantageous further development of the method according to the invention, the measurement signals of all sensors are compared in pairs first with the measurement data of the one and then respectively with the measurement data of the next sensor, after which it is determined which of the sensors has the decalibration. A minimal to slight decalibration of a sensor can then be compensated according to a very advantageous development of the idea, provided that it is known accordingly and does not have a safety-critical influence on the measuring principle of the sensor. This can be done in particular by the software by a recalibration of the decalibrated sensor is made as long as the deviation of its measurement signals is on average smaller than a predetermined limit. As a result, the sensor network can continue to be operated safely and reliably. If there are increasing recalibrations, for example in the region of a sensor during operation, then this is recognized accordingly and can be recorded. This makes it possible, for example, to generate a "warning" and to give the operator of the system with the sensor network, such as a vehicle, the opportunity to have the sensor network checked to prevent more serious errors.

According to a further embodiment of the method according to the invention, it may also be provided to trigger a warning message, and in particular an emergency shutdown, when the mean value of the deviation of the measurement signals is greater than a predetermined limit value or exceeds such longer than a defined maximum predetermined duration. This can be done in any case, regardless of the number of sensors, and regardless of whether the method itself can determine which of the sensors is affected. In an advantageous manner, the size of the limit value is smaller in systems which do not permit recalibration, since they can not detect the affected sensor, than in the case of systems which permit recalibration and therefore can even to a lesser extent compensate for decalibration.

A further advantageous embodiment of the method according to the invention now also provides that the measurement signals are subjected to a preparation by means of a noise filter and / or a Kalman filter before the cross comparison. This makes it possible to suppress or filter out a noise which can not be avoided in principle, or an average value on which the sensor signals settle, quickly and reliably detect. In this way it can be avoided that noise-related extreme values are used in the cross comparison and therefore lead to false errors, readjustments and / or warning messages.

In this case, the method according to the invention can continuously carry out the cross comparison of the measurement signals during operation. It is thus able to detect a possible decalibration and / or malfunction of one of the sensors extremely quickly, in particular faster than it could by the self-diagnosis of a sensor, which is typically done at power and then optionally from time to time. In particular, a gradual change can be detected via the continuous cross-comparison of the sensors, which, unlike the self-diagnosis of the sensors, indicates extremely quickly the (incipient) malfunction of a sensor, eg. B. a self-building on a radar sensor ice layer or the like.

As already mentioned, the method according to the invention can now be used in particular for detecting the surroundings in a mobile system. According to a very advantageous development of the use, it may further be provided that the mobile system uses the environment version for autonomous locomotion. In such mobile systems, such as vehicles, self-propelled logistics systems, or the like, the environment socket is important to ensure safe and reliable movement of the mobile system in space without collisions with objects, other mobile systems, or the like.

In particular, in the case of the mobile system, which is moved autonomously with the aid of the environment version, it may be a vehicle, in particular a passenger vehicle or a commercial vehicle. Especially in such an autonomous vehicle driving application, the safe and reliable operation of the sensors and a safe and reliable assessment of their correct orientation play a crucial role. For example, can cause a significant shift or rotation of the individual sensors, parts of the sensors or their mechanical attachment to each other by temperature differences, as they can occur over a longer period of time on only one side of the vehicle. As a result, for example, stereo cameras can squint and / or the axis of a radar system can be rotated relative to the axis of a camera. In this case, reliable and reliable distance detection is no longer possible. However, since this is a fundamental safety-relevant function for use in autonomous driving, in these cases, the problem must be recognized very quickly on the one hand, and on the other hand responding quickly. For smaller deviations, this can be done by the mentioned recalibration, if it is exactly known due to a larger number of sensors, which sensor is dekalibriert to what extent.

Otherwise, especially when a warning message is triggered according to claim 5 of the method according to the invention, according to a very advantageous development of this use, the autonomous driving can then be stopped. Such a "break off" of autonomous driving means that a user of the vehicle is signaled that he is again responsible, typically after a certain defined period of time as takeover time, to drive the vehicle itself. If the driver does not actively acknowledge the acquisition, an emergency shutdown with a stop of the vehicle may also be provided. Thus, safety-critical situations, due to autonomous driving in any case safely and reliably prevented by an early emergency shutdown of autonomous driving.

A further advantageous embodiment of the method according to the invention and its use results from the exemplary embodiment, which is described in more detail below with reference to the single attached figure.

The sole figure shows the exemplary use of the method according to the invention in a vehicle for autonomous driving.

The figure underlying scenario shows a vehicle 1 , which exemplifies with three sensors 2 . 3 . 4 equipped as environment sensors. This may be, for example, a radar 2 , a mono or stereo camera 3 and a laser scanner or LIDAR 4 act. These sensors 2 . 3 . 4 are to be understood as purely exemplary. Of course, other sensors in the context of the method according to the invention are conceivable. In the embodiment shown here, it should now be so that with the sensors 2 . 3 . 4 equipped vehicle 1 one with 5 designated vehicle in the direction of travel F vorausfährt. The sensors 2 . 3 . 4 capture the rear of this vehicle 5 , which due to the inevitable statistical dispersion within the 6 designated ellipse is located. This is done both by the camera 3 as well as the LIDAR 4 correctly recognized in the embodiment shown here in terms of position and distance. The sensor 2 in the form of the radar, on the other hand, it captures those from the rear of the vehicle 5 outgoing statistical dispersion in the area of 7 designated ellipse. The ellipses 6 . 7 overlap to some extent what with regard to the analysis of the measuring signals of the individual sensors 2 . 3 . 4 helpful. Therefore, the data of each sensor can 2 . 3 . 4 namely the same target object, in this case the rear of the vehicle 5 , be assigned. Now it is that by comparing the position of the ellipses 6 . 7 it can be determined that these are filtered after filtering for noise and unavoidable inaccuracies in the sensors 3 and 4 to match. The mean values of the deviation are therefore permanently approximately zero, with the exception of noise, so that the measuring signals lie on one another. The measuring signal of the sensor 2 , namely the radar, does not correspond to this signal. It can be stated that the statistical dispersion of the measuring signal in the area of the ellipse 7 outside the measuring signals of the other two sensors 3 . 4 lies. This deviation may be due, for example, to a decalibrated axis of the sensor 2 be triggered. As long as it is safe and reliable to ensure that all sensors 2 . 3 . 4 the same target object, namely the rear of the vehicle 5 , capture, and that the deviation of the measuring signals of the sensor 2 If it does not exceed a predetermined tolerance threshold, this can be recalibrated if necessary, ie the orientation of the axis can be corrected electronically by software intervention. This is especially true when using three or more sensors 2 . 3 . 4 possible because it can be deduced typically, which of the sensors 2 . 3 . 4 is de-calibrated. As long as the deviation is correspondingly small, then the recalibration can be carried out. If the deviation increases, for which a corresponding limit value can be established, for example, on the basis of simulations and / or empirical values, a warning message is generated and, if necessary, autonomous driving of the vehicle 1 canceled, because this would not be safe.

The method thus enables a very secure and reliable monitoring of the measurement signals of at least one sensor 2 . 3 . 4 in particular, for autonomous driving, safe and reliable functionality of the sensors 2 . 3 . 4 also to ensure measurement errors, which by the self-diagnosis of the sensors 2 . 3 . 4 can not be recorded. For smaller deviations, a corresponding recalibration is possible, and in the case of a continuous need for recalibration also a warning message can be issued. In the case of a safety-critical deviation of the measuring signals of one of the sensors 2 . 3 . 4 can be exposed to the environment detection and switched off the autonomous driving of the vehicle accordingly, to continue to ensure safety.

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• US 2014/0307247 A1 [0005]
• US 2014/0333468 A1 [0006]

Cited non-patent literature

• M. Gabb, S. Krebs, O. Löhlein and M. Fritzsche, "Probabilistic Interference of Visibility Conditions by Means of Sensor Fusion," in the IEEE Intelligent Vehicles Symposium, 2014 [0004]

Claims (10)

Method for monitoring the measuring signal of a sensor ( 2 . 3 . 4 ) from a sensor network with at least two sensors ( 2 . 3 . 4 ), which have an at least partially overlapping measuring space, and which have the same target object with different physical measuring principles ( 5 ), characterized in that a cross comparison of the measuring signals of the individual sensors ( 2 . 3 . 4 ), wherein the mean values of the deviations of the measuring signals relative to one another are detected, and wherein in the case of a permanent deviation of the mean values from zero to a relative decalibration of one of the sensors ( 2 . 3 . 4 ) is closed. A method according to claim 1, characterized in that the sensor composite at least three sensors ( 2 . 3 . 4 ) with different physical measuring principles. A method according to claim 2, characterized in that the measuring signals of all sensors ( 2 . 3 . 4 ) in pairs first with the measurement signals of the one and then each with the measurement signals of the next sensor ( 3 . 4 . 2 ), after which it is determined which of the sensors ( 2 . 3 . 4 ) has the decalibration. A method according to claim 3, characterized in that a recalibration of the decalibrated sensor ( 2 . 3 . 4 ) is carried out as long as the deviation of its measuring signals is on average smaller than a predetermined limit value. Method according to one of claims 1 to 4, characterized in that a warning message, and in particular an emergency shutdown, is triggered when the mean value of the deviation of the measuring signals is greater than a predetermined limit value. Method according to one of claims 1 to 5, characterized in that the measurement signals are subjected before the cross comparison of a preparation by means of a noise and / or Kalman filter. Method according to one of claims 1 to 6, characterized in that the cross comparison of the measuring signals during operation takes place continuously. Use of the method according to one of claims 1 to 7, for environment detection in a mobile system. Use according to claim 8, characterized in that the mobile system uses the environment detection for autonomous driving. Use according to claim 9, characterized in that in the case of a warning message according to claim 5, the autonomous driving is stopped.

Images