DE102020203965A1 - Processing system and method for redundant processing of input signals - Google Patents

Abstract

The invention relates to a processing system (1a, 1b, 1c, 1d) comprising a first processing unit (4, 4a) and at least one second processing unit (5, 5a, 5b), which are designed for the parallel processing of input signals, the first processing unit ( 4,4a) and the at least one second processing unit (5,5a, 5b) are designed for redundant processing of identical input signals to generate a first and a second output signal, the first output signal passing through the first processing unit (4,4a) and the second output signal are generated by the second processing unit (5, 5a, 5b), a filter unit (2) being provided which sends the identical input signals with a first bandwidth to the first processing unit (4,4a) and with a second bandwidth to the at least second processing unit (5, 5a, 5b), the first and second bandwidths being different. The invention also relates to e in proceedings.

Description

The invention relates to a processing system comprising a first and at least one second processing unit which are designed for parallel processing of input signals, the at least two processing units operating in parallel being designed for redundant processing of identical input signals to generate a first and a second output signal, the first Output signal are generated by the first processing unit and the second output signal by the at least second processing unit. The invention also relates to a method for redundant processing of input signals.

Driving function applications, for example in the autonomous operating mode with a high degree of autonomy, run in normal operation in a multiprocessor system-on-chip on different processing units. The multiprocessor system-on-chip is a system-on-a-chip (SoC) that contains several microprocessors. A multiprocessor system-on-chip is usually intended for embedded applications. The multiple processing units each contain specific functionalities.

To provide fail-safe systems in aircraft, rail or vehicle technology, e.g. braking or restraint systems that work on an electronic basis, two identical processing units, such as a microcomputer, microprocessor or microcontroller, are provided to generate redundancy in order to achieve a high level of safety and reliability . It is assumed here that, in the parallel processing of the algorithms, the redundant processing unit works with the same performance as the primary processing unit.

If one processing unit fails, the other processing unit takes over the arithmetic operation. The heat dissipation is therefore identical.

The number of these processing units and heat losses, as well as the complexity of these systems and the associated error probability represent a major challenge.

the WO 1997 006 487 A1 discloses a microprocessor system for safety-critical regulations, with two synchronously operated central processing units or CPUs, which receive the same input information and process the same program, with read-only memories (ROM) and read-write memories (RAM), with storage locations for test data and with test data generators Comparators that check the output data or output signals of the central units and emit switch-off signals if they do not match, the central units or CPUs being connected via separate bus systems to the read-only memories and to the read-write memories and to input and output units, and where the bus systems are interconnected by driver stages which enable the two central units to read and process the pending data, ie the data available in the two bus systems, including the test data and commands.

the EP 2693278 B1 discloses a method for operating an image processing control device, an output date being determined from at least one input data by means of an algorithm provided for image processing, the input data being an image from a camera, whereby it is checked whether the output data is in a first, not all possible output data Group is included, according to which a second determination of the output date takes place only in the case of an output data contained in the first group to check the output data, the first group, which does not contain all possible output data, describes an intervention by an actuator and / or for an intervention by an actuator includes leading output data as well as safety-critical output data.

It is the object of the present invention to develop a fail-safe processing system in such a way that a high level of security and reliability can be achieved at lower costs and / or a high overall utilization of the fail-safe processing system.

The object is achieved by a processing system with the features of claim 1 and a method with the features of claim 15.

The object is achieved by a processing system comprising a first processing unit and at least one second processing unit, which are designed to process input signals in parallel, the at least two processing units operating in parallel being designed to redundantly process identical input signals to generate a first output signal and a second output signal , the first output signal being generated by the first processing unit and the second output signal being generated by the at least second processing unit, a filter unit being provided which transmits the identical input signals with a first bandwidth to the first processing unit and with a second bandwidth to the at least second processing unit transmitted, the first and second bandwidths being different.

The input signals are, for example, sensor data / actuator data that are generated by sensors / actuators.

The processing system according to the invention transmits the input signals with different bandwidths to the processing units, which each apply an identical function (application software) to the input signals. By using different input signals, different inhomogeneous processing units can be used.

By using such a redundant processing unit, the second output signals can be used exclusively for plausibility checking and validation of the first output signals, the determined first output signals being used to form an action / for further processing.

As a result, the second processing unit can be designed with less computing capacity than the first processing unit. As a result, redundant processing of the input signals can also be achieved with an inexpensive processing system.

If the processing units are arranged on a carrier plate, for example a semiconductor substrate, a reduction can be achieved not only in the computing capacity but also in the installation space required. When arranged on a common carrier plate, the processing system according to the invention also requires less heat during processing, which leads to the use of smaller cooling components.

As a result, a processing system according to the invention with redundant processing units is specified, which is inexpensive and also generates less heat. This results in less heat loss.

The processing system according to the invention does not require application software balancing, since all functions run, for example, on the first processing unit and the at least second processing unit has the identical application software as the first processing unit and is only used for plausibility checks and validation. This reduces the complexity of the system, which also significantly reduces the scope of testing and validation of the basic algorithms in the pre-series stage or when the product is launched on the market.

The input signals are preferably transmitted to the second processing unit with a smaller bandwidth. If there are more than two processing units, the third processing unit, for example, contains the input signals with a smaller bandwidth than the second.

The filter unit is preferably designed to filter the input signals with different weightings. This makes it very easy to achieve the different bandwidths. There is no data compression, neither lossless nor lossy, only a filtered data provision, based, for example, on an image or pixel representation of the input signals. Intensity values that are too high are preferably not used in processing because they do not carry enough information.

A comparison unit is preferably provided for comparing the first and second output signals, the bandwidth being dynamically adapted if the second output signal deviates from the first output signal. The lower bandwidth can be iteratively adapted to the higher bandwidth. Furthermore, the filter unit is preferably provided in order not to undertake any data compression when filtering the input signals with different weightings. This means that no computing power is required.

The filter unit is preferably designed to transmit the input signals to the first processing unit unfiltered or with a first degree of filtering and to the at least second processing unit with a second degree of filtering. In the case of a first degree of filtering or unfiltered input signals, the bandwidth is preferably higher than in the case of a second degree of filtering.

The input signals are preferably transmitted to the first processing unit with full bandwidth, the output signals of the second and possibly the subsequent processing units only being used for plausibility checking or validation of the output signals of the first processing unit. The actions / further processing can, for example, be determined solely by the output signals of the first processing unit.

Furthermore, the filter unit is preferably designed to reduce the second filter degree to a second new filter degree and to calculate a new second output signal if the second output signal deviates from the first output signal. Alternatively, if the second output signal deviates from the first output signal, the filter unit is designed to iteratively reduce the second filter level to a second new filter level to calculate a new second output signal until the first output signal matches the second output signal or until the second new filter degree corresponds to the first filter degree or a non-filtering. A reduction in the degree of filtering is to be equated with an increase in the bandwidth.

As a result, a quasi-parallel processing system is achieved which interacts dynamically.

The first and the at least one second processing unit preferably have an identical or smaller computing capacity for processing the input signals. Furthermore, the at least one second processing unit can be designed to make unused computing capacity internally available to the vehicle. As a result, for example, non-safety applications such as multimedia applications can be executed on this unused computing capacity within the E / E architecture of the vehicle.

Alternatively, the first processing unit has a first computing capacity for processing the input signals and the at least one second processing unit has a second computing capacity for processing the input signals, the second computing capacity being less than the first computing capacity. Above all, this configuration enables costs to be saved that would otherwise be incurred with redundant processing.

The input signals are preferably designed as images, the filter unit being designed to transmit the unfiltered image or to filter the image with the first filter degree, the first filter degree being designed as a first image segmentation and the first image segmentation to be transmitted as an input signal and wherein the filter unit is further designed to filter the image with the second filter degree, the second filter degree being designed as a second image segmentation and to transmit the second image segmentation as an input signal, the second image segmentation being greater than the first image segmentation or the unfiltered one Image. However, there is no image compression here either. Image segmentation is used here as the filter level. A reduction in the bandwidth can thus be achieved in a simple manner.

Furthermore, the input signals are preferably designed as images which have pixel and bit depth, the first processing unit being designed to determine the first output signal in relation to a pixel and / or bit depth of the image, the second processing unit for this purpose is designed to determine the second output signal in relation to the same pixel and / or bit depth of the image, so that corresponding first and second output signals are generated for all existing pixels and / or bit depths of the image, and the comparison unit is designed to do so to compare the generated output signals pixel by pixel and / or bit by bit.

In this case, the bandwidth can be formed by a different bit depth. The first processing unit can receive the input signals unfiltered, i.e. in full bandwidth. For example, the second processing unit receives the input signals in 50% filtering. This means that the resolution of the bit depths is halved, which corresponds to approximately half the payload of the image data. The output signals generated by the first and second processing unit are compared, for example, pixel by pixel. If there is no match, the resolution, that is to say the bit depth, can be successively adapted to the second processing unit. In this way, for example, errors can be determined in a spatially resolving manner. In comparison to this, in conventional parallel-processed systems, the fault detection in a spatially resolving manner is not possible, since either the first or the second processing unit has a fault.

The object is also achieved by a method for redundant processing of input signals, which is designed to be carried out on a processing system as described above.

Further features, properties and advantages of the present invention emerge from the following description with reference to the accompanying figures. Variations therefrom can be derived by those skilled in the art without departing from the scope of protection of the invention as defined by the following patent claims.

• 1: eine erste Ausgestaltung eines erfindungsgemäßen Verarbeitungssystems,
• 2: eine weitere Ausgestaltung eines erfindungsgemäßen Verarbeitungssystems,
• 3: eine weitere Ausgestaltung eines erfindungsgemäßen Verarbeitungssystems,
• 4: eine weitere Ausgestaltung eines erfindungsgemäßen Verarbeitungssystems,
• 5: Ausgangssignale in Bezug auf die jeweilige Verarbeitungseinheit,
• 6: ein Bild mit einem zu erkennenden Objekt, durch Pixel dargestellt,
• 7: das Bild von 6 mit unterschiedlicher Bandbreite,
• 8: eine Häufigkeitstabelle.

The figures show schematically:

• 1 : a first embodiment of a processing system according to the invention,
• 2 : a further embodiment of a processing system according to the invention,
• 3 : a further embodiment of a processing system according to the invention,
• 4th : a further embodiment of a processing system according to the invention,
• 5 : Output signals in relation to the respective processing unit,
• 6th : an image with an object to be recognized, represented by pixels,
• 7th : the picture of 6th with different bandwidth,
• 8th : a frequency table.

1 shows a first embodiment of a processing system according to the invention 1a .

This has a filter unit 2 which is designed to receive input signals. The input signals can come from sensors, for example 3 / Actuators are generated.

In this case, the - not encompassing - image sensors / lidar sensors / radar sensors can also be acoustic sensors, acceleration sensors or a combination thereof.

The filter unit 2 receives the identical input signals.

The filter unit 2 can be designed depending on the type of input signals received, for example as an FPGA module (Field Programmable Gate Array).

The filter unit 2 is designed to send identical input signals to a first processing unit 4th and a second processing unit 5 to submit. The processing units can be microprocessors or controllers, in particular a so-called performance controller, namely a GPU (graphics processing unit), etc. The GPU itself can consist of multicore computing units.

Here is the filter unit 2 also designed to send the identical input signals in different bandwidths to the first processing unit 4th and the second processing unit 5 to submit. For this purpose, the input signals are filtered with different weightings. However, the filtering does not perform any data compression; neither lossless nor lossy. Only a filtered data provision is provided on the basis of, for example, an image or pixel representation.

So preferably receives the first processing unit 4th the input signals unfiltered and the second processing unit 5 for example with a 50% filtering as a second degree of filtering. This means that the first processing unit 4th receives the input signals in full bandwidth and the resolution of the input signals for the second processing unit 5 are almost halved. The input signals are so to speak through the filter unit 2 "Quasi" cut off.

The first processing unit 4th and the second processing unit 5 execute identical functions (application software), preferably safety processing steps for safety applications, in parallel and thus generate a first and a second output signal, the second processing unit 5 The input signal is the input signal with a reduced bandwidth. This means that redundant processing is identical with regard to the information on which the input signal is based.

The second processing unit is due to the reduced bandwidth 5 with less computing capacity than the first processing unit 4th configured, that is, that the first processing unit 4th has a first computing capacity for processing the input signals and the second processing unit 5 has a second computing capacity, wherein the second computing capacity is less than the first computing capacity.

Above all, this configuration enables costs to be saved that would otherwise be incurred with redundant processing.

The first output signal generated in this way is then sent by the first processing unit 4th and the second output signal from the second processing unit 5 to a comparison unit 6th transmitted. In this, the output signals are checked for plausibility and validated if they match.

The output signals indicate whether an event, for example an object, was detected or not.

If the first output signal is “positive”, that is, an event has been recognized, and the second output signal is “negative”, that is, if the two output signals do not match, then it is assumed that an error is holistic or local on the first processing unit 4th or the second processing unit 5 is present.

With such a deviation of the second output signal from the first output signal, the bandwidth, here of the second processing unit 5 , dynamic ( D. ) can be adjusted by reducing the second degree of filtering and thus the bandwidth of the second processing unit 5 transmitted input signals is increased, that is, for example, the resolution is increased.

This increase can be carried out successively until the result of the first output signal and the second output signal match.

If they match, an action can be taken, for example, depending on the type and information of the output signal A. / Further processing A be made by the vehicle or the driver.
The second degree of filtering can be selected manually, for example as a function of the safety applications that are set or operated or executed by means of the first output signal.

Alternatively, the second degree of filtering can be set by an artificial neural network.

Likewise, the successive increase in the bandwidth in the event of a mismatch can have been set in advance.

2 shows a further embodiment of a processing system 1b according to the invention.

As in 1 the filter unit 2 which is designed to receive input signals. The input signals can come from sensors, for example 3 / Actuators are generated.

The filter unit 2 receives the identical input signals.

The filter unit 2 is designed to send the identical input signals in different bandwidths to the first processing unit 4th and the second processing unit 5a to submit. For this purpose, the input signals are filtered with different weightings.

So preferably receives the first processing unit 4th the input signals unfiltered and the second processing unit 5a for example with a 50% filtering as a second degree of filtering.

Alternatively, a first degree of filtering can be provided, for example if the output signals are made available to a non-safety application. Here, however, the second degree of filtering is higher than the first degree of filtering; that is, the first processing unit 4th receives the input signals with a higher bandwidth than the second processing unit 5a .

The first processing unit 4th and the second processing unit 5a execute identical functions (application software), preferably safety processing steps for safety applications, in parallel and thus generate a first and a second output signal, the second processing unit 5a The input signal is the input signal with a reduced bandwidth.

In the processing system 1b however, assign the first processing unit 4th and the second processing unit 5a an identical computing capacity for processing the input signals. Due to the lower bandwidth of the input signals that are sent to the second processing unit 5a are transmitted, however, there is free computing capacity in the second processing unit 5a available.

This can for example be made available to the vehicle internally for processing data / executing functions, for example in order to execute other non-safety applications, for example multimedia applications.

Alternatively, it can also be used to process externally recorded environmental data. If a negative event is detected for, for example, another road user, for example a vehicle with a high level of autonomy, a warning can be generated wirelessly to the other road user.

If the first output signal is “positive”, that is, an event has been recognized, and the second output signal is “negative”, that is, if the two output signals do not match, it is assumed that an error is holistic or local on the first processing unit 4th or the second processing unit 5a is present.

In the event of such a deviation of the second output signal from the first output signal, the bandwidth can be adjusted dynamically ( D. ), in which the second degree of filtering is reduced. This increases the bandwidth of the second processing unit 5a transmitted input signals increased, that is, for example, increased the resolution.

If, for example, the output signal is a recognition of an object and the input signal is an image that is represented by the input signals pixels and bit depths of the image, the frequency with which the object is in the individual bit depths and / or pixels of the can be determined Image by the first processing unit 4th and with the same object in the individual bit depths and / or pixels of the image by the second processing unit 5a was perceived / recognized. If these frequencies differ too greatly, the bandwidth can be increased dynamically.

This increase can be carried out successively until the result of the first output signal and the second output signal match.
The second degree of filtering can be selected manually, for example as a function of the safety applications that are set or operated or executed by means of the first and second output signals. Alternatively, the second degree of filtering can be set by an artificial neural network.

3 shows a further embodiment of a processing system according to the invention 1c . This has a filter unit 2 on, which is designed here as an FGPA module.

The filter unit 2 receives as input signals images that have a bit depth and pixels from sensors 3 , which are designed here as an image camera. The filter unit is here 2 also designed to send the identical input signals in different bandwidths to the first processing unit 4a and the second processing unit 5b to submit. At least the image is sent to the second processing unit 5b transmitted with reduced bandwidth.

In order to obtain a reduced bandwidth, the images are in the filter unit 2 segmented, i.e. the picture frame in which the image is evaluated is of different size or only certain areas on the image are evaluated. Image compression still does not take place. Therefore, no additional computing power is required.

So preferably receives the first processing unit 4a the image completely, i.e. unsegmented with full bandwidth unfiltered and the second processing unit 5b with a predetermined image segmentation.

The second processing unit can also be used here 5b be designed with less computing capacity, which leads to a cheaper overall system or with the same computing capacity, which can be provided internally for processing data or executing a preferably non-safety application.

For example, the output signals can indicate whether an event, for example an object / vehicle, was detected in the image or not.

If the first output signal is “positive”, that is, an event has been recognized, and the second output signal is “negative”, that is, if the two output signals do not match, then it is assumed that an error is holistic or local on the first processing unit 4a or the second processing unit 5b is present.

In the event of such a deviation of the second output signal from the first output signal, the bandwidth can be adjusted dynamically ( D. ), in which the image for the second processing unit 5b segmented less. This lower segmentation increases the bandwidth of the input signals for the second processing unit 5b elevated.

For example, if the output signal is a recognition of an object and the input signal is an image, the frequency can be determined, how often an object is the first output signal and how often the same object is the second output signal, for example based on the individual bit depths and / or pixels of the image were perceived / recognized. If these frequencies differ too greatly, the bandwidth can be increased dynamically.
This increase can be carried out successively until the result of the first output signal and the second output signal match.

If the results match, further action / further processing can take place depending on the output signal A. be made.

For example, if the vehicle in front is recognized, braking can be carried out by the vehicle.

4th shows a further embodiment of a processing system according to the invention 1d .

This indicates the filter unit 2 on, which is designed here as an FGPA module. Furthermore, there is also an image camera as a sensor 3 intended for the generation of images. The image information of the camera for image recordings is provided for the series status in YUV format with a bit depth of Y = 12bit, U = 8bit and V = 8bit and in pixels.

The first processing unit 4a receives the images (image data) as input signals unfiltered, i.e. in full bandwidth.

The second processing unit also receives 5b the images (image data) are filtered as input signals, i.e. with a reduced bandwidth. The second filter level is set so that the input signals are transmitted with 60% of the full bandwidth. This means that the resolution in the YUV color space is almost halved, which corresponds to roughly half the payload of the image data.

There is also a third processing unit 7th provided with a third degree of filtering, which is higher than the second degree of filtering.

The third degree of filtering is set so that the input signals are transmitted with 30% of the full bandwidth.

There is also a further x processing unit 8th provided, where x stands for any number of processing units. This has / have an x-th filter degree which is higher (that is, in each case a lower bandwidth) than the previous (x-1) filter degree. The x filter degree becomes like this set so that the input signals are transmitted with x% of the full bandwidth.

The recognition of an object in the input signals is assumed here as the output signal.

The first processing unit 4a , the second processing unit 5b , the third processing unit 7th and the nth processing unit 8th determine the first output signal, the second output signal, the third output signal and the x-th output signal by using the same function, i.e. the same identical application software. Using the same application software reduces the complexity of the system.

The use of the identical application software also has the advantage that no application software balancing is necessary since, for example, all driver assistance systems - driving functions or autonomy - driving functions are carried out by the first processing unit 4a depend.

If the output signals are available, they are compared with one another and checked for plausibility. The plausibility check also means a validation of the output signals. The recognition or non-recognition in the first processing unit 4a , the second processing unit 5b the third processing unit 7th and the nth processing unit 8th can be recorded in a frequency table.

5 shows such a recognition of the output signals in relation to the respective processing unit by way of example. This is done to the first processing unit 4a 100% of the image transmitted (unfiltered, full bandwidth) to the second processing unit 5b only 60% of the image (filtered, reduced bandwidth) to the third processing unit 7th only 30% of the image (filtered, reduced bandwidth) and to the umpteenth processing unit 8th at least, namely x% of the image. Since the image is transmitted with pixels and bit depths, the first processing unit 4a all pixels and bit depths are transmitted to the nth processing unit 8th least bit depths (or pixels).

Then the frequency with which the object in the image is determined in relation to the pixels and the bit depth by the first processing unit 4a , the second processing unit 5b , the third processing unit 7th and the nth processing unit 8th is recognized.

Also shows 5 the frequency M of the recognition as a function of the YUV format with a bit depth of Y = 3bit, U = 3bit and V = 3bit and the first processing unit 4a , the second processing unit 5b , the third processing unit 7th and the nth processing unit 8th at.

6th shows an image with an object to be recognized, which is represented by pixels. Some of these pixels have the object to be recognized here, while others do not.

It can thus be examined pixel by pixel whether the respective processing unit 4a , 5b , 7,8 detects the object in the same pixels or not. As the processing units 4a , 5b , 7,8 use the same processing function (application software), but not the same bandwidth, despite the identical processing function, different output signals can occur in relation to the individual pixels.

If the change in bandwidth is not in the bit depths, but in a different segmentation, it can be, for example, that the nth processing unit 8th in the case of strong segmentation, no longer has any or only few positive matches.

7th shows the picture of 6th heavily segmented, which is sent to the umpteenth processing unit 8th is transmitted.

8th shows the frequency of 5 as a frequency table for the first processing unit 4a , the second processing unit 5b and the third processing unit 7th at. According to the frequency table, whether the first output signal of the first processing unit 4a with the output signal of the second processing unit 5b and the third processing unit 7th corresponds in the individual bit depths and / or pixels, that is to say, for example, whether or not an object has been recognized, a plausibility check is carried out. Is the first output signal related to a bit / pixel of the first processing unit 4a “Positive”, and for example the results from the other processing units 5b , 7th "Negative", it is assumed that an error is holistic or local on one of the processing units 4a , 5b , 7th is available (row 2, row 6).

The degree of filtering of the second processing unit is successively iterative 5b reduced, that is, there is a larger bandwidth or the input signals are transmitted with a larger bandwidth, up to a value until the output signal corresponds to the first output signal in terms of bit depth / pixel. If there is no match, the third processing unit will also 7th less filtered. The procedure is for the other processing units 8th etc. same.

If the first output signal is equal to “negative” in relation to the individual bit and / or pixel, but the output signals of the second processing unit 5a and the third processing unit 7th "Positive" (row 6 in the table of 8th ), the error can be determined spatially resolved. But it can also be a defective first processing unit 4a exist or the images are incorrectly transmitted.

Are the output signals at all processing units 4a , 5b , 7th "Positive", at least that is how the third processing unit works 7th with too high a resolution and therefore requires too many resources for image processing. This is in turn 7th the table in 8th shown. This means, among other things, an increased heat emission, which leads to a shorter service life of the processing system 1d can lead, preferably when all processing units 4a , 5b , 7th are arranged in the vicinity.
The probabilities of an error calculated value in a processing system 1d with x-times executed processing units is thus 1 / x. This means that there are different probabilities or an unequal probability distribution.

In a conventional parallel-processed system, the fault detection is not possible in a spatially resolving manner, since the processing units are only viewed as a whole and it is therefore only recognized that one of the processing units or both have an error. According to the invention, it is possible to detect the errors in an n-fold GPU arrangement in a spatially resolving manner. As a result, the error can be compensated for in a targeted manner with different filtering levels without the overall system being impaired as a result of the error.

The degree of filtering can be reduced for images by lowering the image segmentation or by using a higher resolution.

The first processing unit 4th can have a first computing capacity for processing the input signals, the second processing unit 5b a second computing capacity and the third processing unit 7th and the nth processing unit 8th have an x-th computing capacity for processing the input signals, the second computing capacity being less than the first computing capacity, the third being less than the second computing capacity and the x-th computing capacity being less than the (x-1) computing capacity.

This can save costs.

Alternatively, however, the first processing unit 4a and the second processing unit 5b and the third processing unit 7th and the nth processing unit 8th have an identical computing capacity for processing the input signals. Due to the lower bandwidth of the input signals which are sent to the second processing unit 5a are transmitted, but there is free computing capacity available. The same applies to the third processing unit 7th and the nth processing unit 8th . This free computing capacity can be made available to the vehicle for other use.

The processing units are able to determine the (individual) computing power in advance and to determine the filter degrees in advance or to control the individual GPUs accordingly. Thus, the filter degrees are not configured reactively to an error or to the error frequency, but the configuration is preventive.

One advantage of the present method when determining errors is that the calculations are based on different probabilities or with unequal probability distributions.

List of reference symbols

1a-1d

Processing system

2

Filter unit

3rd

Sensors

4,4a

first processing unit

5,5a, 5b

second processing unit

6th

Comparison unit

7th

third processing unit

8th

x-th processing unit 8

A.

action

D.

Dynamic adjustment of the bandwidth

QUOTES INCLUDED IN THE DESCRIPTION

This list of the documents listed by the applicant was 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.

Patent literature cited

• WO 1997006487 A1 [0006]
• EP 2693278 B1 [0007]

Claims (15)

Processing system (1a, 1b, 1c, 1d) comprising a first processing unit (4,4a) and at least one second processing unit (5,5a, 5b), which are designed for the parallel processing of input signals, wherein the first processing unit (4,4a) and the at least one second processing unit (5, 5a, 5b) is designed for redundant processing of identical input signals to generate a first and a second output signal, the first output signal by the first processing unit (4,4a) and the second output signal by the second processing unit (5, 5a, 5b), characterized in that a filter unit (2) is provided which sends the identical input signals with a first bandwidth to the first processing unit (4, 4a) and with a second bandwidth to the at least second processing unit ( 5, 5a, 5b), the first and second bandwidths being different. Processing system (1a, 1b, 1c, 1d) according to Claim 1 , characterized in that the filter unit (2) is designed to filter the input signals with different weightings. Processing system (1a, 1b, 1c, 1d) according to Claim 2 , characterized in that the filter unit (2) is provided for the purpose of not undertaking any data compression when filtering the input signals with different weightings. Processing system (1a, 1b, 1c, 1d) according to one of the preceding claims, characterized in that a comparison unit (6) is provided for comparing the first output signals and second output signals, the bandwidth being dynamically adjusted if the second output signal differs from the first output signal will. Processing system (1a, 1b, 1c, 1d) according to Claim 4 , characterized in that the filter unit (2) is designed to send the input signals to the first processing unit (4, 4a) unfiltered or with a first filter level and to the at least one second processing unit (5, 5a, 5b) with a second filter level to transfer. Processing system (1a, 1b, 1c, 1d) according to Claim 5 , characterized in that with a first degree of filtering or unfiltered input signals, the bandwidth is higher than with a second degree of filtering. Processing system (1a, 1b, 1c, 1d) according to Claim 6 , characterized in that the filter unit (2) is designed to reduce the second filter degree to a second new filter degree and to calculate a new second output signal if the second output signal deviates from the first output signal. Processing system (1a, 1b, 1c, 1d) according to Claim 6 , characterized in that the filter unit (2) is designed to iteratively reduce the second filter level to a second new filter level when the second output signal deviates from the first output signal and to calculate a new second output signal until the first output signal matches the second Output signal matches or until the second new filter degree corresponds to the first filter degree or a non-filtering. Processing system (1a, 1b, 1c, 1d) according to one of the preceding claims, characterized in that the first processing unit (4, 4a) and the at least one second processing unit (5, 5a, 5b) have identical computing capacity for processing the input signals. Processing system (1a, 1b, 1c, 1d) according to Claim 9 , characterized in that the at least one second processing unit (5, 5a, 5b) is designed to make unused computing capacity internally available to the vehicle. Processing system (1a, 1b, 1c, 1d) according to Claim 9 or 10 , characterized in that the second processing unit (5, 5a, 5b) is designed to use unused computing capacity for non-safety applications. Processing system (1a, 1b, 1c, 1d) according to one of the preceding Claims 1 until 8th , characterized in that the first processing unit (4,4a) has a first computing capacity for processing the input signals and the at least one second processing unit (5,5a, 5b) has a second computing capacity for processing the input signals, the second computing capacity being less than that first computing capacity is. Processing system (1a, 1b, 1c, 1d) according to one of the preceding claims, characterized in that the input signals are designed as images, the filter unit (2) being designed to transmit the unfiltered image or to filter the image with the first Carry out filtering degree, wherein the first filtering degree is designed as a first image segmentation and to transmit the first image segmentation as an input signal and wherein the filter unit (2) is also designed to filter the image with the second filtering degree, the second filtering degree as a second Image segmentation is formed and the second image segmentation as an input signal transmit, wherein the second image segmentation is greater than the first image segmentation or the unfiltered image Processing system (1a, 1b, 1c, 1d) according to one of the preceding claims, characterized in that the input signals are designed as images which have pixel and bit depth, the first processing unit (4,4a) being designed to do the to determine the first output signal with respect to a pixel and / or bit depth of the image, the second processing unit (5, 5a, 5b) being designed to assign the second output signal with respect to the same pixel and / or bit depth of the image so that corresponding first and second output signals are generated for all existing pixels and / or bit depths of the image, and the comparison unit is designed to compare the generated output signals with one another pixel and / or bit by bit. Method for redundant processing of input signals, which is designed to be carried out on a processing system (1a, 1b, 1c, 1d) according to one of the preceding claims.

Images