DE102018126826A1 - Method for determining a confidence value for a detection quality of a road marking boundary, driver assistance system and motor vehicle - Google Patents

Abstract

The invention relates to a method for determining a confidence value (C) for a detection quality of a road marking boundary (K) of a road (9) for a motor vehicle (1):
- Determining a lane marking limit (K) of a detected lane (9) as a function of the detected lane marking (10, 11) by means of an electronic computing device (12) of the driver assistance system (2);
- generating at least one parameter value (V1 to V7) of a parameter (P1 to P7) characterizing the road marking boundary (K) as a function of the detection quality by means of a predefined confidence model (M) by means of the electronic computing device (12);
- Classifying the characterizing parameter (p1 to P7) depending on the generated parameter value (V1 to V7) by means of a predetermined membership function (Z) for the characterizing parameter (P1 to P7) according to the predetermined confidence model (M); and
- Determining the confidence value (C) as a function of the classified characterizing parameter (P1 to P7) by means of the electronic computing device (12).
The invention further relates to a computer program product, a driver assistance system (2) and a motor vehicle (1).

Description

The invention relates to a method for determining a confidence value for a detection quality of a road marking boundary for a driver assistance system of a motor vehicle. The invention further relates to a driver assistance system for carrying out the method and a motor vehicle with a driver assistance system.

The recognition of road markings is of great importance for many different driver assistance systems. In addition to autonomous driving systems and lane keeping systems, an example of this is, for example, a system for recognizing parking spaces or for detecting a parking space. Road markings are usually recognized at the edge of an image of the surroundings of the motor vehicle. For example, the image of the surroundings is received from a camera of the motor vehicle or several cameras of the motor vehicle are composed from raw images. The road marking in the image can then be recognized by known algorithms of an image evaluation. For this purpose, for example, image gradients are formed over different areas of the image and a road marking is then recognized if the image gradient corresponds to a predetermined pattern.

For the driver assistance system to function correctly, it is necessary for road markings to be recognized reliably and safely. For example, it is of crucial importance for a lane departure warning system that the lane marking can be reliably recognized. In particular, due to, for example, missing lane markings or due to damage to the lane marking, it can happen that a lane marking can be misinterpreted, which can lead to critical situations in road traffic. In a fully autonomous operation of the motor vehicle in particular, it is of crucial importance that the road marking can be reliably determined or recognized.

The object of the present invention is to provide a method, a driver assistance system and a motor vehicle, by means of which a detection quality of a roadway can be reliably determined.

This object is achieved by a method, a driver assistance system and a motor vehicle in accordance with the independent patent claims.

One aspect of the invention relates to a method for determining a confidence value for a detection quality of a road marking boundary for a driver assistance system of a motor vehicle. A road marking in an environment of the motor vehicle is detected by means of a detection device of the driver assistance system. The road marking limit is determined as a function of the detected road marking by means of an electronic computing device of the driver assistance system. At least one parameter value of a parameter characterizing the road marking limit is generated as a function of the detection quality by means of a predetermined confidence model by means of the electronic computing device. The characterizing parameter is classified depending on the generated parameter value by means of a predetermined membership function for the characterizing parameter according to the predetermined confidence model. The confidence value is determined as a function of the classified characterizing parameter by means of the electronic computing device.

This enables a quality of the detection of the road marking for a driver of the motor vehicle or, for example, for a further driver assistance system for the partially autonomous, in particular for the autonomous, driving operation of the motor vehicle. In other words, the confidence value is used to describe how well or how badly the road marking boundary has been detected. The reliability of the recognition of the lane marking can thereby be determined. In particular, it can then be provided, for example, if, for example, a low detection quality with a low confidence value is output that a driver of the motor vehicle should take over the driving task in partially autonomous driving operation or in fully autonomous driving operation, so that a critical situation due to poor detection of the lane marking is prevented can. Furthermore, if the quality of the detection is low, the driver can be given a corresponding warning signal, so that the driver is warned that the detection quality of the lane marking is correspondingly low. If a high detection quality with a high confidence value is determined, this can lead to the autonomous driving operation of the motor vehicle being able to be carried out with a high degree of reliability.

In particular, a position, a bend or an orientation of the road marking can be determined. In particular, the corresponding properties of the road marking can then be assessed by means of the confidence value. In particular, it is provided that the roadway is defined by the road markings. For this purpose, for example, the lane can have a lane marking on a left and a right side of the lane. The lane markings on the left and on the right side can differ, but can also have the same design. For example, a solid road marking can be formed on a right side of the road, while a dashed road marking can be formed on a left side of the road.

In particular due to missing lane markings or due to damage to the lane marking, errors can occur when the lane marking is detected. This in turn can lead to critical situations in road traffic.

By evaluating the lane marking on the basis of the detection quality, it can thus be prevented that, in the event of a poor detection of the lane marking, appropriate decisions are made which could lead to a critical situation. In particular, in the case of poor detection, for example, the driving task can then be handed over to a driver of the motor vehicle. Critical situations in road traffic can thus be prevented.

In particular, it can be provided according to the invention that a large amount of data is evaluated when the lane marking is detected and the confidence value is output accordingly. Using this confidence value, a further driver assistance system that can process the confidence value can then make further decisions, for example. This means that the processing in the further driver assistance system can be carried out more easily, since this can only continue to work on the basis of the confidence value without having to process the large amount of data further. As a result, a decision of the further driver assistance system, for example with regard to taking over driving tasks for the driver, can be carried out in a simplified and faster manner.

Furthermore, it can be provided that corresponding predetermined rules and regulations regarding the detection quality are taken into account by means of the driver assistance system. For example, it can be taken into account that the confidence value is output low when the driver assistance system is started, since the data recorded so far are still too low to be able to assess the detection quality as high. As a result, the confidence value is output in accordance with predefined criteria. The confidence value for the lane marking can thus be output individually and depending on the situation, so that this can lead to increased safety in road traffic.

In particular, the confidence value represents how reliably and precisely the shape of the lane marking was recognized for transmission to another driver assistance system. For example, if the recognized lane marking limit should be very close to the real lane marking limit, the confidence value can be classified as very high.For example, a low confidence value of the system should be output because the shape of the generated lane marking limit does not correspond to the real lane marking limit this leads to the further driver assistance system carrying out a corresponding reaction, for example a transfer of the driving task.

For example, an overall length of the road marking can be regarded as the characterizing parameter. The distance between the respective end points of the detected lane marking can be viewed as the total length. In particular, it can be provided that the higher the parameter value of the total length, the better the detection quality. An accumulated length of the lane markings can also be regarded as the characterizing parameter. In particular, it is the sum of the individual markings. Here, too, it can be seen that the higher this value, the higher the detection quality. For example, a residual average value can be regarded as the characterizing parameter. This can represent how well the lane marking boundary fits into the confidence model. The lower the better. For example, the average value of the road marking quality can be regarded as the characterizing parameter. In particular, the mean value of the detection quality can be used for each lane marking within the different detection steps. The higher this value, the higher the detection quality. For example, anything that represents the ratio of the lane marking that fits into the selected model for the lane marking limit can be regarded as a characterizing parameter. The higher this value, the higher the detection quality. For example, a detection of the road can serve as a characteristic parameter. In other words, there should be one The lane marking limit has been recorded in order to output the confidence value. If this has been recorded accordingly, the confidence value will increase accordingly; if no lane marking has been recorded, the confidence value will decrease accordingly. Furthermore, a counter can be used as the characterizing parameter which represents how long a corresponding lane marking has been recorded. The higher this value, the higher the detection quality. The road marking can thus be detected by means of different characterizing parameters, the characterizing parameters in particular having a different influence on the determination of the confidence value. As a result, the lane marking can be recorded redundantly with different characterizing parameters and the confidence value can be determined. This leads to an increased output quality of the confidence value. Furthermore, this leads to increased safety in road traffic.

It can preferably be provided that the method according to the invention is carried out on a single electronic computing device. It is also possible that the different method steps can be carried out on different electronic computing devices.

According to an advantageous embodiment, the characterizing parameter is classified into at least two, in particular three, classes depending on the parameter value. Corresponding detection quality can be output simply and nevertheless reliably, in particular by classification into two, in particular three, classes. For example, it can be stated that the characterizing parameter was recorded with a high quality or was recorded with a low quality. This means that the characterizing parameter is classified in a simple manner and in a resource-saving manner and can be passed on to the further driver assistance system. In particular, the recording quality can then be evaluated in a simple manner so that it can reliably prevent a critical situation in road traffic. The number of classes is to be understood as an example only. The characterizing parameters can also be classified into more than three classes.

It has also proven to be advantageous if a first class describes a low confidence with regard to the characterizing parameter, with a second class an average confidence with regard to the characterizing parameter and with a third class a high confidence with regard to the characterizing parameter as a function of the given confidence model becomes. For example, it should be provided that the values of the confidence value lie between 0 and 1, so 0 can describe a low confidence, ie a low detection quality, and 1 can describe a high detection quality. For example, it can be provided that between 0 and 0.3 the first class with the low confidence is described as the confidence value, between 0.3 and 0.6 the middle class is described with the middle confidence value and with the values 0.6 to 1 the third class is described with the high confidence value. This makes it possible for the characterizing parameter to be classified as a function of the respectively divided class with the respectively divided parameter as a function of the respective parameter value, as a result of which the detection quality of the road marking boundary can be reliably determined.

Alternatively, it can be provided that a first class describes a high confidence with regard to the characterizing parameter, with a second class an average confidence with respect to the characterizing parameter and with a third class a low confidence with regard to the characterizing parameter as a function of the specified confidence model.

In a further advantageous embodiment, a multiplicity of differing parameters characterizing the road marking boundary is determined and a respective parameter value of a respective characterizing parameter is generated as a function of the detection quality and a respective characterizing parameter is dependent on the respective parameter value by means of a respective membership function for the respective characterizing parameter of the given confidence model and the confidence value is determined as a function of the large number of classified characterizing parameters. In particular, since the lane marking boundary can have differently characterizing parameters, a detailed evaluation regarding the detection quality of the lane marking boundary can be carried out by detecting the different characterizing parameters. In particular, it is then provided that a respective parameter value is recorded for each characterizing parameter. In particular, these are different characterizing parameters. In particular, the respective parameter values can then also differ. Each characterizing parameter has an associated membership function within the confidence model. Each of the plurality of characterizing parameters then becomes dependent on the respective membership function classified accordingly. The characterizing parameters can each have different classes. It is provided that the confidence value is then determined as a function of the characterizing parameters. The confidence value can thus be determined redundantly and with a large number of different characterizing parameters. This leads to increased security with regard to the detection quality. The detection quality of the road marking boundary with respect to the confidence value can thus be reliably determined.

It has also proven to be advantageous if a respective predetermined parameter weighting value is assigned to a respective characterizing parameter and the confidence value is determined as a function of the respective weighted classified parameters. In other words, it can be provided that the different characterizing parameters are assigned a different weighting within the confidence model. This weighting is specified on the basis of the parameter weighting values. For example, it can be provided that a recorded total length of the lane marking is assigned a high weighting value compared to a temporal length of the detection of the lane marking. In other words, it can be provided that, for example, although the parameter value of the time recording is classified as high and, for example, the overall length is assigned a low parameter value, the confidence value can nevertheless be classified as low, since the parameter value of the total length of the road marking is given greater weight can. In particular, the weighting values can be determined beforehand. In particular, these can be dependent on corresponding field tests, so that a corresponding weighting can be assigned to the respective characterizing parameters. This leads to an improved determination of the detection quality of the road marking boundary, as a result of which increased safety in road traffic can be achieved.

Furthermore, it has proven to be advantageous if at least one control function is generated with a large number of characterizing parameters and the confidence value is determined as a function of the control function. In particular, it can be regarded as a control function that, for example, a low confidence is to be output when the driver assistance system is started. In particular, this can be determined via the different characterizing parameters within the control function. In particular, it can be provided that the control function has the plurality of characterizing parameters, which are in particular different, and the confidence value can then be determined on the basis of this control function. The characterizing parameters within the control function can have different parameter weighting values for the control function, so that the importance of the individual characterizing parameters is determined as a function of the respective control function. As a result, the confidence value for the detection quality of the road marking can be reliably output as a function of the control function.

In other words, the control functions can be determined from a combination of one or more classes from one or more membership functions.

In a further advantageous embodiment, a large number of different control functions are generated with a large number of characterizing parameters and a respective control weight value is assigned to a respective control function and a control function value is determined as a function of the respective control weighting values and the confidence value is determined as a function of the respective control function value. In other words, a variety of control functions can be generated. The respective control function can then in turn be determined from different characterizing parameters. In particular, it can be provided that a characterizing parameter is also taken into account in several control functions, for example. In other words, a characterizing parameter can influence different control functions. For example, a control function can be determined in such a way that when the system is started, it should have a low confidence value. Furthermore, it can be provided as a control function, for example, that a low confidence value should be output in the case of a low number of road markings. Furthermore, it can be provided, for example, if the lane markings have been recorded precisely that the confidence value is rated accordingly high. This means that a large number of different control functions can influence the confidence value for the acquisition quality. Furthermore, it can be provided that a corresponding control weighting value is assigned to the respective control function. In other words, a respective predefined control function can have different importance for the influence for determining the confidence value. In particular, it can be provided that these standard weighting values are specified and have arisen, for example, on the basis of field tests. For example, the control function that a low confidence value is output when the system is started may have a rule weighting value of 0.05, while, for example, the control function that the road marking is of low or high quality is assigned a medium confidence value and this is assigned a rule weighting value of, for example, 1. In other words, the influence of the second control function mentioned is higher than the influence of the first control function mentioned. As a result, the individual rules are weighted depending on the field tests, which means that the confidence value for the detection quality of the lane marking can be reliably output.

It is also advantageous if a value of a rule influence for determining the confidence value is determined by the sum of the respective parameter values of the respective rule. In other words, the control influence of the individual parameter values of the characterizing parameters is determined.

der Konfidenzwert bestimmt werden kann, wobei der Konfidenzwert (Confidence) Werte zwischen 0 und 100 einnehmen kann, wobei 0 einem minimalen Konfidenzwert und 100 einem maximalen Konfidenzwert entsprechen kann.It can further be provided that the confidence value is determined as the average of the control function values (Ruleweight) over the control influences (Rulelnfluence). In particular, the control influence can lie between the values 0 and 1. For example, it can then be provided that the formula: $$\mathrm{C.}Onfidence=100{\displaystyle \sum \left(RuleweiGHt\ast Rule\mathrm{I.}nfluence\right)/{\displaystyle \sum Rule\mathrm{I.}nfluence}}$$

the confidence value can be determined, whereby the confidence value can have values between 0 and 100, where 0 can correspond to a minimum confidence value and 100 can correspond to a maximum confidence value.

It has also proven to be advantageous if the characterizing parameter is used to determine a detection performance and / or a model setting and / or a detection reliability and / or a model selection and / or a time consistency of the road marking boundary. For example, a total length of the road marking can be viewed as the detection performance. The distance between the respective end points of the detected lane marking can be viewed as the total length. In particular, it can be provided that the higher the parameter value of the total length, the better the detection quality. Furthermore, an accumulated length of the lane markings can be regarded as the detection performance. In particular, it is the sum of the individual markings. Here, too, it can be seen that the higher this value, the higher the detection quality. For example, a residual mean value can be viewed as a model setting. This can represent how well the lane marking boundary fits into the confidence model. The lower the better. For example, the mean value of the road marking quality can be regarded as the detection reliability. In particular, the mean value of the detection quality can be used for each lane marking within the different detection steps. The higher this value, the higher the detection quality. For example, anything that represents the ratio of the lane marking that fits into the selected model for the lane marking limit can be viewed as a model selection. The higher this value, the higher the detection quality. A detection of the roadway can serve as time consistency, for example. In other words, a road marking boundary should have been detected at all in order to output the confidence value. If this has been recorded accordingly, the confidence value will increase accordingly; if no lane marking has been recorded, the confidence value will decrease accordingly. Furthermore, a counter can be used as the time consistency, which represents how long a corresponding lane marking has been recorded. The higher this value, the higher the detection quality. The road marking can thus be detected by means of different characterizing parameters, the characterizing parameters in particular having a different influence on the determination of the confidence value. As a result, the lane marking can be recorded redundantly with different characterizing parameters and the confidence value can be determined. This leads to an increased output quality of the confidence value. Furthermore, this leads to increased safety in road traffic.

It has also proven to be advantageous if the confidence value is determined by means of a fuzzy logic of the electronic computing device. Using fuzzy logic, a characteristic of a property as a degree of membership can be numerically recorded and thus a fuzziness can be precisely modeled. The fuzzy logic is based on the fuzzy sets. The set is not defined as before by the objects that are elements of this set, but by the degree to which they belong to this set. This is done in particular through the membership functions, which assign a numerical value as a degree of membership to each element. The new set operations introduced in this way define the operations as an associated logic calculus that allows the modeling of indifference processes. In other words, it can be provided that, depending on the parameter value, the characterizing parameter can be classified into two classes, for example. For example, it can be provided that with the characterizing parameters of both the first class and the second class can be assigned to a parameter value of 0.3. As a result, the blurring of the characterizing parameters can be taken into account in a simple manner. This leads to a reduction in misinterpretation, since blurring is also taken into account. The confidence value with regard to the acquisition quality can thus be reliably determined.

According to a further advantageous embodiment, the predefined confidence model can be predefined as a function of regulations and / or predefinable rules. These regulations and / or prescribable rules can originate from the definition of the system or can be determined on the basis of field tests and behavior. For example, it can be provided that the system fulfills the requirement that the driver assistance system outputs a low confidence value when changing the lane of the motor vehicle. Furthermore, it could also be provided, for example, that if the length of the road markings is estimated to be very short, the driver assistance system also outputs a predetermined low confidence value. This can be based on field tests, for example, since it was found within the field test that the accumulated lengths of the road markings can be classified as very short and therefore have a low reliability, so that the driver assistance system can have a low confidence value.

It has also proven to be advantageous if the membership function is determined as a function of the frequency distribution of the false positive and the correct positive and the false negative predicted values for the characterizing parameter. This enables a binary classifier to be provided. To evaluate a classifier, you have to use it in a number of cases in which you have at least retrospective knowledge of the “true” class of the respective objects. An example of such a case is a field test to determine whether it fits into a class. This represents a classifier in order to be able to classify the characterizing parameter in a first class, a second class or, for example, a third class. True positive describes that the characterizing parameter falls into the class and that this was correctly determined. False negative means that the characterizing parameter falls into the class, but has been divided into the wrong class. False positive means that the characterizing parameter does not fall into the class, but the characterizing parameter was nevertheless classified incorrectly.

As a result, the respective classification of the characterizing parameter can be carried out in a simple manner using a binary classifier. This enables the classification to be carried out in a simple manner and the confidence value to be determined.

A further aspect of the invention relates to a driver assistance system with at least one detection device and an electronic computing device, the driver assistance system being designed to carry out the method according to the preceding aspect or an advantageous embodiment thereof. In particular, the method or an advantageous embodiment thereof is carried out on the driver assistance system.

A still further aspect of the invention relates to a computer program product with program code means which are stored in a computer-readable medium in order to carry out the method according to the previous aspect or an advantageous embodiment thereof when the computer program product is processed on a processor of an electronic computing device.

Yet another aspect of the invention relates to a motor vehicle with a driver assistance system according to the previous aspect. The motor vehicle is designed in particular as a passenger car.

Advantageous embodiments of the method are to be regarded as advantageous embodiments of the driver assistance system and of the motor vehicle. For this purpose, the driver assistance system and the motor vehicle have physical features which enable the method to be carried out and an advantageous embodiment thereof.

Further features of the invention result from the claims, the figures and the description of the figures. The features and combinations of features mentioned above in the description and the features and combinations of features mentioned below in the description of the figures and / or shown alone in the figures can be used not only in the combination specified but also in other combinations or on their own, without the scope of the invention leave. Embodiments of the invention are thus also to be regarded as encompassed and disclosed, which are not explicit in the figures are shown and explained, but emerge from the separated combinations of features from the explanations and can be generated. Versions and combinations of features are also to be regarded as disclosed, which therefore do not have all the features of an originally formulated independent claim. In addition, versions and combinations of features, in particular those disclosed by the above-described embodiments, are to be regarded as disclosed, which go beyond or differ from the combinations of features set out in the back references of the claims.

Exemplary embodiments of the invention are explained in more detail below with the aid of schematic drawings.

• 1 eine schematische Draufsicht auf eine Ausführungsform eines Kraftfahrzeugs mit einer Ausführungsform des Fahrerassistenzsystems;
• 2 ein schematisches Ablaufdiagramm für die Durchführung einer Ausführungsform des Verfahrens; und
• 3 ein schematisches Diagramm einer Ausführungsform einer Zugehörigkeitsfunktion.

Show:

• 1 a schematic plan view of an embodiment of a motor vehicle with an embodiment of the driver assistance system;
• 2nd a schematic flow diagram for performing an embodiment of the method; and
• 3rd a schematic diagram of an embodiment of a membership function.

In the figures, identical or functionally identical elements are provided with the same reference symbols.

1 shows a motor vehicle 1 according to an embodiment of the present invention in a plan view. The car 1 is designed in the present case as a passenger car. The car 1 has a driver assistance system 2nd on. The driver assistance system 2nd again has at least one image processing device 3rd on, for example, by an electronic control unit of the motor vehicle 1 can be formed. In addition, the driver assistance system 2nd at least one camera 4th as a detection device for the driver assistance system 2nd on. In the present exemplary embodiment, the driver assistance system has 2nd for example four cameras 4th on that distributed to the motor vehicle 1 are arranged. However, only two cameras can be used, for example 4th or three cameras 4th or more than four cameras 4th be provided. One of the cameras is present 4th in a rear area 5 arranged one of the cameras 4th is in a front area 7 of the motor vehicle 1 arranged and the remaining two cameras 4th are in a respective page area 6 , in particular in an area of the side mirrors. The number and arrangement of cameras 4th of the camera system 2nd is to be understood here purely as an example.

With the cameras 4th can be an environment 8th of the motor vehicle 1 be recorded. The four cameras 4th are preferably constructed identically. In particular, a sequence of images with the cameras 4th be provided which the environment 8th describe. These pictures can be taken from the cameras 4th to the image processing device 3rd be transmitted. The driver assistance system 2nd thus serves to support the driver of the motor vehicle 1 when driving the motor vehicle 1 . The driver assistance system 2nd can be, for example, an ACC (Adaptive Cruise Control System) or a parking assistance system or another system.

The 1 shows the motor vehicle 1 on a lane 9 . The roadway 9 has a first lane marking in the present exemplary embodiment 10th , which in particular a solid road marking 10th and a variety of non-solid lane markings 11 on, in the present example four of these road markings 11 are displayed. The road markings 10th , 11 each have road edges K on which in particular a respective lane marking 10th , 11 limit horizontally. The road markings 10th , 11 each have a lane marking width H. Furthermore, in particular in the present case by the lane marking 11 , a lane line L are formed, which by optical stringing the respective lane markings 11 arises.

It is envisaged that a confidence value C. ( 2nd ) for a detection quality of the lane marking limit K for the driver assistance system 2nd of the motor vehicle 1 is determined. It becomes the lane marking 10th , 11 in the neighborhood 8th of the motor vehicle 1 by means of the detection device, in the present case in particular using the cameras 4th of the driver assistance system 2nd , detected. It becomes the lane marking boundary K depending on the detected lane marking 10th , 11 by means of the electronic computing device 12th of the driver assistance system 2nd certainly. There will be at least one parameter value V1 to V7 ( 2nd ) one the road marking boundary K characterizing parameter P1 to P7 ( 2nd ) depending on the acquisition quality using a predefined confidence model M ( 2nd ) by means of the electronic computing device 12th certainly. It becomes the characterizing parameter P1 to P7 depending on the generated parameter value V1 to V7 by means of a predetermined membership function Z ( 3rd ) for the characterizing parameters P1 to P7 according to the given confidence model M classified. It becomes the confidence value C. depending on the classified characterizing parameter P1 to P7 by means of the electronic computing device 12th certainly.

This enables a quality of the detection of the lane marking K for a driver of the motor vehicle 1 or for example for a further driver assistance system for the partially autonomous, in particular for the autonomous, driving operation of the motor vehicle 1 is issued. In other words, using the confidence value C. described how good or how bad the lane marking limit K has been recorded. This can improve the reliability of the recognition of the lane marking 10th , 11 be determined. In particular, it can then be provided, for example, if, for example, a low detection quality with a low confidence value C. are output that, for example, a driver of the motor vehicle in partially autonomous driving mode or in fully autonomous driving mode 1 should take over the driving task, so that a critical situation due to poor detection of the lane marking 10th , 11 can be prevented. Furthermore, if the detection quality is low, the driver can be given a corresponding warning signal, so that the driver is warned that the detection quality of the road marking 10th , 11 is correspondingly low. Should be a high acquisition quality with a high confidence value C. can be determined, this can lead to a high reliability of the autonomous driving of the motor vehicle 1 can be carried out.

In particular, a position, a bend or an orientation of the lane marking 10th , 11 be determined. In particular, using the confidence value C. then the corresponding properties of the road marking 10th , 11 be rated. In particular, it is provided that through the road markings 10th , 11 the roadway 9 is defined. For example, the roadway 9 on a left and a right side of the road 9 one lane marking each 10th , 11 exhibit. The road markings 10th , 11 on the left and on the right side can differ, but can also have the same design. For example, a solid road marking can be found on a right side of the road 10th be formed while a dashed road marking on a left side of the road 11 can be trained.

Especially due to missing lane markings 10th , 11 or due to damage to the lane marking 10th , 11 It can happen that when the lane marking is detected 10th , 11 errors can occur. This in turn can lead to critical situations in road traffic.

By evaluating the lane marking 10th , 11 on the basis of the detection quality, it can thus be prevented that the lane marking is poorly recorded 10th , 11 appropriate decisions are made that could lead to a critical situation. In particular, in the case of poor detection, for example, the driving task can be directed to a driver of the motor vehicle 1 be handed over. Critical situations in road traffic can thus be prevented.

In particular, it can be provided according to the invention that a large amount of data is acquired when the lane marking is detected 10th , 11 is assessed and the confidence value C. is issued accordingly. By means of this confidence value C. another driver assistance system can then make further decisions, for example. This means that the processing in the further driver assistance system can be carried out more easily, since this is only based on the confidence value C. can continue to work without having to process the large amount of data. As a result, a decision of the further driver assistance system, for example with regard to taking over driving tasks for the driver, can be carried out more easily and quickly.

In particular, the confidence value represents C. how reliable and precise the shape of the road marking 10th , 11 has been recognized for transmission to another driver assistance system. For example, the recognized lane marking limit K in shape very close to the real lane marking boundary K can be the confidence value C. classified as very high should, for example, have a low confidence value C. of the system are output because of the shape of the lane marking boundary generated K not the real lane marking limit K corresponds, this can lead to the further driver assistance system carrying out a corresponding reaction, for example a transfer of the driving task.

For example, as a characterizing parameter P1 to P7 a total length of the lane marking 10th , 11 be considered. The total length can be the distance between the respective end points of the detected road marking 10th , 11 be considered. In particular, can be provided be that the higher the parameter value V1 to V7 the total length, the better the detection quality is. Furthermore, as a characterizing parameter P1 to P7 an accumulated length of the lane markings 10th , 11 be considered. In particular, it is the sum of the individual markings. Here, too, it can be seen that the higher this value, the higher the detection quality. For example, as a characterizing parameter P1 to P7 a residual average can be considered. This can represent how good the lane marking limit is K into the confidence model M fits. The lower the better. For example, as a characterizing parameter P1 to P7 the mean value of the lane marking quality can be viewed. In particular, this allows the mean value of the detection quality within the different detection steps for each lane marking 10th , 11 be used. The higher this value, the higher the detection quality. As a characteristic parameter P1 to P7 For example, everything can be viewed, which is the ratio of the lane marking 10th , 11 , which fits into the selected model, for the road marking boundary K represents. The higher this value, the higher the detection quality. As a characteristic parameter P1 to P7 can, for example, record the road 9 serve. In other words, there should be a road marking boundary at all K have been captured to the confidence value C. to spend. If this has been recorded accordingly, the confidence value C. should increase accordingly, no lane marking 10th , 11 have been recorded, the confidence value is accordingly C. sink. Furthermore, as a characterizing parameter P1 to P7 a counter can be used which represents how long a corresponding lane marking 10th , 11 has been recorded. The higher this value, the higher the detection quality. Thus, by means of different characterizing parameters P1 to P7 the lane marking 10th , 11 are recorded, with the characterizing parameters P1 to P7 especially different influence on the determination of the confidence value C. can take. This allows redundancy with different characterizing parameters P1 to P7 the lane marking 10th , 11 be captured and the confidence value C. be determined. This leads to an increased output quality of the confidence value C. . Furthermore, this leads to increased safety in road traffic.

It can preferably be provided that the method according to the invention on a single electronic computing device 12th is carried out. It is also possible that the different method steps on different electronic computing devices 12th can be carried out.

2nd schematically shows a flow chart of an embodiment of the method. In step S1 is using the camera 4th the lane marking 10th , 11 detected. In step S2 becomes the road marking boundary K certainly. In step S3 becomes a parameter V1 to V7 of the at least one characterizing parameter P1 to P7 certainly.

In particular, it can be provided that the driver assistance system 2nd a fuzzy logic 13 by means of which the confidence value C. can be determined.

In particular, it is provided that a plurality of different, the road marking boundary K characterizing parameters P1 to P7 is determined and a respective parameter value V1 to V7 of a respective characterizing parameter P1 to P7 is generated depending on the acquisition quality and a respective parameter P1 to P7 depending on the respective parameter value V1 to V7 by means of a respective membership function Z for the respective parameter P1 to P7 using the given confidence model M is classified and the confidence value C. depending on the large number of classified parameters P1 to P7 is determined. Furthermore, it can be provided that a respective predetermined parameter weighting value is assigned to a respective characterizing parameter P 1 to P7 and the confidence value C. depending on the respective weighted classified parameters P1 to P7 is determined.

Especially since the lane marking limit K different characterizing parameters P1 to P7 can have, by means of the detection of the different characterizing parameters P1 to P7 a detailed evaluation regarding the detection quality of the road marking boundary K be performed. In particular, it is then provided that for each characterizing parameter P1 to P7 a respective parameter value V1 to V7 is recorded. In particular, these are different characterizing parameters P1 to P7 . In particular, the respective parameter values can then also change V1 to V7 differentiate. Each characterizing parameter P1 to P7 has an associated membership function Z within the confidence model M on. It then becomes each of the multitude of characterizing parameters P1 to P7 classified according to the respective membership function Z accordingly. The characterizing parameters P1 to P7 different classes CL1 , CL2 , CL3 ( 3rd ) exhibit. It is provided that depending on the characterizing parameters P1 to P7 the confidence value C. is determined. It can thus be redundant and with a multitude of different ones characterizing parameters P1 to P7 reliably the confidence value C. be determined. This leads to increased security with regard to the detection quality. The detection quality of the road marking boundary can thus be reliably K regarding the confidence value C. be determined.

In particular, the different characterizing parameters P1 to P7 a different weighting within the confidence model M is assigned. This weighting is specified on the basis of the parameter weighting values. For example, it can be provided that a detected total length of the road marking 10th , 11 a high weighting value compared to a time length of the detection of the lane marking 10th , 11 is assigned. In other words, it can be provided that, for example, although the parameter value V1 to V7 the time acquisition is classified as high, and for example the overall length is a low parameter value V1 to V7 the confidence value can still be assigned C. are classified as low because of the parameter value V1 to V7 the total length of the lane marking 10th , 11 a higher weight can be given. In particular, the weighting values can be determined beforehand. In particular, these can be dependent on corresponding field tests, so that the respective characterizing parameter P1 to P7 an appropriate weighting can be assigned. This leads to an improved determination of the detection quality of the road marking boundary K , whereby an increased safety in road traffic can be realized.

Furthermore, it can be provided that at least one control function R1 , R2 , R3 with a variety of characterizing parameters P1 to P7 is generated and the confidence value C. depending on the control function R1 , R2 , R3 is determined. In particular, it can be provided that a large number of different control functions R1 , R2 , R3 with a variety of characterizing parameters P1 to P7 is generated and a respective control function R1 , R2 , R3 a respective rule weighting value is assigned and a control function value is determined depending on the respective rule weighting values and the confidence value C. is determined depending on the respective control function value.

As a control function R1 , R2 , R3 can be viewed in particular that, for example, when the driver assistance system starts 2nd a low confidence level should be issued. In particular, this can be done using the different characterizing parameters P1 to P7 within the control function R1 , R2 , R3 be determined. In particular, it can be provided that the control function R1 , R2 , R3 the multitude of characterizing parameters P1 to P7 , which are in particular different, and based on this control function R1 , R2 , R3 then the confidence value C. can be determined. The characterizing parameters P1 to P7 within the control function R1 , R2 , R3 can have different parameter weighting values for the control function R1 , R2 , R3 so that the importance of the individual characterizing parameters P1 to P7 depending on the respective control function R1 , R2 , R3 is determined. This can reliably depend on the control function R1 , R2 , R3 the confidence value C. for the detection quality of the lane marking 10th , 11 be issued.

The respective control function R1 , R2 , R3 can then again from different characterizing parameters P1 to P7 be determined. In particular, it can be provided that a characterizing parameter P1 to P7 for example also in several control functions R1 , R2 , R3 is also taken into account. In other words, a characterizing parameter P1 to P7 Influence on different control functions R1 , R2 , R3 to have. For example, a control function R1 , R2 , R3 be determined such that when the system is started, it has a low confidence value C. should have. Furthermore, for example, as a control function R1 , R2 , R3 be provided that with a low number of lane markings 10th , 11 a low confidence level C. should be spent. Furthermore, the lane markings should be provided, for example 10th , 11 that the confidence level C. is classified accordingly high. It can be used with a variety of different control functions R1 , R2 , R3 Influence on the confidence value C. be taken for the acquisition quality. Furthermore, it can be provided that the respective control function R1 , R2 , R3 a corresponding standard weighting value is assigned. In other words, a given predetermined control function R1 , R2 , R3 different importance for the influence to determine the confidence value C. exhibit. In particular, it can be provided that these standard weighting values are specified and have arisen, for example, on the basis of field tests. For example, the control function R1 , R2 , R3 that when the system starts up a low confidence value C. is output, have a rule weighting value of 0.05, for example during the rule function R1 , R2 , R3 that the road marking 10th , 11 is of low or high quality, medium confidence C. is assigned and this is assigned a rule weighting value of, for example, 1. In other words, the influence of the second control function mentioned R1 , R2 , R3 higher than the influence of the control function mentioned first R1 , R2 , R3 . This leads to the fact that depending on the field tests individual rules are weighted, which means that the confidence value is reliable C. for the detection quality of the lane marking 10th , 11 can be spent.

In particular shows 2nd that, for example, a first control function R1 a first characterizing parameter P1 with a parameter value V1 , a second characterizing parameter P2 with a second parameter value V2 and a third characterizing parameter P3 with a third parameter value V3 is assigned. Furthermore, the 2nd that a second control function R2 a fourth characterizing parameter P4 with a fourth parameter value V4 and a fifth characterizing parameter P5 with a fifth parameter value V5 is assigned. A third control function R3 can, for example, as in the present case, be the first characterizing parameter P1 and the fifth characterizing parameter P5 and a sixth characterizing parameter P6 with a sixth parameter value V6 and a seventh characterizing parameter P7 with a seventh parameter value V7 be assigned.

In particular, this shows that a respective control function R1 , R2 , R3 about different characterizing parameters P1 to P7 can be defined. In particular, however, the individual characterizing parameters P1 to P7 also several times in the control functions R1 , R2 , R3 be taken into account. In particular, the respective characterizing parameters can then P1 to P7 have different parameter weighting values, the different parameter weighting values in turn being within the respective rule R1 , R2 , R3 can distinguish.

durchgeführt werden, wobei der Konfidenzwert C Werte zwischen 0 und 100 einnehmen kann, wobei 0 einem minimalen Konfidenzwert C und 100 einem maximalen Konfidenzwert C entsprechen kann.Furthermore, it can be provided that a value of a control influence for the determination of the confidence value C. (Confidence) by the sum of the respective parameter values P1 to P7 the respective control function R1 , R2 , R3 is determined. In particular, the confidence value C. be determined as the average of the control function values (Ruleweight) over the control influences (Rulelnfluence). For example, using the formula: $$\mathrm{C.}Onfidence=100{\displaystyle \sum \left(RuleweiGHt\ast Rule\mathrm{I.}nfluence\right)/{\displaystyle \sum Rule\mathrm{I.}nfluence}}$$

be performed, taking the confidence value C. Can take values between 0 and 100, where 0 is a minimum confidence value C. and 100 a maximum confidence value C. can correspond.

Furthermore, it can in particular be provided that by means of the characterizing parameter P1 to P7 a detection performance and / or a model setting and / or a detection reliability and / or a model selection and / or a time consistency of the lane marking limit K be determined. The acquisition performance and / or the model setting and / or the acquisition reliability and / or the model selection and / or the time consistency can be determined using the different characterizing parameters P1 to P7 to be discribed.

Furthermore, the 2nd that the given confidence model M depending on regulations L and / or predeterminable rules T. These regulations L and / or predefinable rules T can originate from the definition of the system or can be determined on the basis of field tests and behavior. For example, it can be provided that the system fulfills the requirement that when the motor vehicle changes lanes 1 the driver assistance system 2nd a low confidence level C. issues. Furthermore, it could also be provided, for example, that if the length of the road markings 10th , 11 the driver assistance system is considered to be very short 2nd also a predetermined low confidence value C. issues. This can be based on field tests, for example, since it was found within the field test that the accumulated lengths of the road markings 10th , 11 can be classified as very short and therefore have a low reliability, so that the driver assistance system 2nd a low confidence level C. can have.

3rd shows schematically a diagram of an embodiment of the membership function Z. In particular shows 3rd that a blur of the characterizing parameters P1 to P7 is also taken into account.

Furthermore shows 3rd that in particular the characterizing parameter P1 to P7 depending on the parameter value V1 to V7 in at least two, especially three, classes CL1 , CL2 and CL3 is classified. In particular, for example, by means of the first class CL1 a low confidence in the characterizing parameter P1 to P7 , by means of the second class CL2 an average confidence regarding the characterizing parameter P1 to P7 and by means of the third class CL3 a high confidence regarding the characterizing parameter P1 to P7 depending on the given confidence model M to be discribed.

In particular, the 3rd that by means of fuzzy logic 13 the blurring can be realized with. In the present embodiment, the class CL1 with the low confidence value C. defined from 0 to 0.4. The second class CL2 with the mean confidence value is defined from 0 to 0.8. The third class with the high confidence value C. is defined from 0.7 to 1. In particular, depending on the respective parameter value V1 to V7 the characterizing parameter P1 to P7 in the respective class CL1 , CL2 , CL3 be classified.

In particular, it is provided that the membership function Z as a function of the frequency distribution of the false positive and the correct positive and the false negative predictive values for the characterizing parameter P1 to P7 is determined. This enables a binary classifier to be provided. To evaluate a classifier, you have to use it in a number of cases where, at least in retrospect, you have knowledge of the "true" class CL1 , CL2 , CL3 of the respective objects. An example of such a case is a field trial to determine if he is in a class CL1 , CL2 , CL3 fits. This represents a classifier to the characterizing parameter P1 to P7 in a first class CL1 , a second class CL2 or for example third grade CL3 to be able to classify. True positive describes that the characterizing parameter P1 to P7 in the class CL1 , CL2 , CL3 falls and this has also been correctly determined. False negative means that the characterizing parameter P1 to P7 in the class CL1 , CL2 , CL3 falls, but in the wrong class CL1 , CL2 , CL3 was divided. False positive means that the characterizing parameter P1 to P7 not in the class CL1 , CL2 , CL3 falls, but the characterizing parameter P1 to P7 was classified incorrectly.

As a result, the respective classification of the characterizing parameter can be easily done using a binary classifier P1 to P7 be performed. This enables the classification and the confidence value to be carried out in a simple manner C. be determined.

Overall, the figures show a method for determining a detection quality of the road marking boundary K a driver assistance system 2nd .

Claims (15)

Method for determining a confidence value (C) for a detection quality of a road marking boundary (K) of a road (9) for a motor vehicle (1) by means of a driver assistance system (2), with the steps: - Detecting a road marking (10, 11) of a road (9) in an environment (8) of the motor vehicle (1) by means of a detection device of the driver assistance system (2); - Determining a road marking limit (K) as a function of the detected road marking (10, 11) by means of an electronic computing device (12) of the driver assistance system (2); - generating at least one parameter value (V1 to V7) of a parameter (P1 to P7) characterizing the road marking boundary (K) as a function of the detection quality by means of a predefined confidence model (M) by means of the electronic computing device (12); - Classifying the characterizing parameter (p1 to P7) depending on the generated parameter value (V1 to V7) by means of a predetermined membership function (Z) for the characterizing parameter (P1 to P7) according to the predetermined confidence model (M); and - Determining the confidence value (C) as a function of the classified characterizing parameter (P1 to P7) by means of the electronic computing device (12). Procedure according to Claim 1 , characterized in that the characterizing parameter (P1 to P7) is classified depending on the parameter value (V1 to V7) in at least two, in particular in three, classes (CL1, CL2, CL3). Procedure according to Claim 2 , characterized in that by means of a first class (CL1) a low confidence regarding the characterizing parameter (P1 to P7), by means of a second class (CL2) an average confidence regarding the characterizing parameter (P1 to P7) and by means of a third class ( CL3) a high confidence with regard to the characterizing parameter (P1 to P7) depending on the given confidence model (M) is described. Method according to one of the preceding claims, characterized in that a plurality of different parameters (P1 to P7) characterizing the road marking limit (K) is determined and a respective parameter value (V1 to V7) of a respective characterizing parameter (P1 to P7) in Dependency of the detection quality is generated and a respective characterizing parameter (P1 to P7) is classified depending on the respective parameter value (v1 to V7) by means of a respective membership function (Z) for the respective characterizing parameter (P1 to P7) by means of the predetermined confidence model (M) and the confidence value (C) is determined as a function of the large number of classified characterizing parameters (P1 to P7). Procedure according to Claim 4 , characterized in that a respective predetermined parameter weighting value is assigned to a respective characterizing parameter (P1 to P7) and the confidence value (C) is determined as a function of the respective weighted classified characterizing parameters (P1 to P7). Procedure according to Claim 5 , characterized in that at least one control function (R1, R2, R3) is generated with a large number of characterizing parameters (P1 to P7) and the confidence value (C) is determined as a function of the control function (R1, R2, R3). Procedure according to Claim 6 , characterized in that a large number of different control functions (R1, R2, R3) is generated with a large number of characterizing parameters (P1 to P7) and a respective control weight value is assigned to a respective control function (R1, R2, R3) and a control function value in Dependency of the respective control weighting values is determined and the confidence value (C) is determined as a function of the respective control function value. Procedure according to one of the Claims 6 or 7 , characterized in that a value of a control influence for determining the confidence value (C) is determined by the sum of the respective parameter values (V1 to V7) of the respective control function (R1, R2, R3). Procedure according to Claim 8 , characterized in that the confidence value (C) is determined as the average of the control function values over the control influences. Method according to one of the preceding claims, characterized in that the characterizing parameter (P1 to P7) is used to determine a detection performance and / or a model setting and / or a detection reliability and / or a model selection and / or a time consistency of the road marking limit (K). Method according to one of the preceding claims, characterized in that the confidence value (C) is determined by means of a fuzzy logic (13) of the electronic computing device (12). Method according to one of the preceding claims, characterized in that the predefined confidence model (M) is predefined as a function of regulations (L) and / or predefinable rules (T). Method according to one of the preceding claims, characterized in that the membership function (Z) is determined as a function of the frequency distribution of the false positive and the true positive and the false negative prediction values for the characterizing parameter (P1 to P7). Driver assistance system (2) with at least one detection device and an electronic computing device (12), the driver assistance system (12) for performing the method according to one of the Claims 1 to 13 is trained. Motor vehicle (1) with a driver assistance system (2) Claim 14 .

Images