DE102020131995A1 - Validator for a position determination - Google Patents

Abstract

A validator for N independent pieces of information indicative of a geographic location has N inputs for the information and M outputs. Using machine learning, the validator is trained to provide a predetermined signal at one of the outputs precisely when one of the pieces of information matches a predetermined position hypothesis less than a predetermined level. Furthermore, the validator is trained to avoid an incorrect missing signal on two or more outputs.

Description

The invention relates to determining a geographic position of a vehicle. In particular, the invention relates to safeguarding the position determination against possible errors.

A vehicle can have a driving assistant that is set up to influence a longitudinal or lateral control of the vehicle. For example, a lane assistant can be set up to keep the vehicle laterally between lane markings that can be scanned by a camera and automatically recognized.

Many driving assistants require knowledge of an exact position of the vehicle. The position can be determined in the longitudinal and/or transverse direction and expressed with respect to a predetermined reference point. An absolute geographic position may be determined with respect to a predetermined geodetic reference system such as the WGS 84, for example. A relative position of the vehicle can be specified, for example, in the transverse direction with respect to a recognized lane marking.

The determination of the position of the vehicle is usually subject to a number of errors and inaccuracies. For example, sensors provide noisy and/or corrupted information, or may occasionally fail altogether. Different measurement conditions or complex processing heuristics lead to different degrees of accuracy or reliability. If the vehicle is controlled based on such a determined position, the safety of the vehicle or an occupant may be compromised.

In DE 10 2019 133 316 A1 it has been proposed to separately validate scans from a single information source that provides data to determine position. A validator can be used to do this, which determines how well the readings from this sensor match the readings from the other sensors. A probability that a given position is wrong, even though all validators confirm the samples assigned to them, equals the product of the probabilities of error of each individual validator.

A limitation of this technique is that the individual validators must be statistically independent of one another in order to allow theoretical or formal proof of a predetermined maximum error probability. More complex methods that carry out several interlinked validations cannot therefore be used.

An object on which the invention is based is to enable improved position determination for a vehicle. The invention solves the problem by means of the subject matter of the independent claims. Subclaims reflect preferred embodiments.

A first aspect of the present invention relates to a validator for N mutually independent pieces of information indicative of a geographical position. The validator has N inputs for the information and M outputs and is trained by means of machine learning to provide a predetermined signal at one of the outputs precisely when one of the pieces of information matches a predetermined position hypothesis worse than a predetermined degree. Furthermore, the validator is trained to avoid an incorrect missing signal on two or more outputs.

The validator is configured to determine whether or not the information fits the position hypothesis sufficiently well. Here a signal is provided which indicates insufficiently good matching information; If, on the other hand, the information matches the position hypothesis, no signal is output. An inverse behavior, in which a signal is output that indicates sufficiently good matching information, with no signal indicating a larger deviation of the information from the position hypothesis, can be implemented in a corresponding manner.

The signals provided at the various outputs can be made largely statistically independent of one another thanks to the additional learning objective of avoiding multiple "false positive" determinations. In this way, each output can be viewed as a separate validator whose error probability can be determined individually. An overall error probability of the validator can be determined as a product of the individual error probabilities. In this way, it can be proven that the validator correctly determines information that does not match in a sufficient number of cases.

The validator can further comprise a combination device connected to the M outputs for determining an integrity of a position determinable on the basis of the information. The integrity may be determined based on an overall error probability of the validator. The combiner can provide a signal indicating whether the position hypothesis is likely to be correct or not.

The position hypothesis can be determined on the basis of the determined information and the validator can be used to determine whether the position hypothesis is correct with sufficient certainty or whether it has a position error that is smaller than a predetermined position error. The performance of the validator can be sufficient to use it to validate a specific geographic position of a vehicle, especially if the vehicle is controlled depending on the determined position. Should the validator indicate that the position hypothesis is not correct with a predetermined certainty, the position hypothesis can be rejected or re-determined based on only some of the determined information. Signals at the validator outputs can indicate information to be discarded or an untrustworthy source for the information.

When determining the total error probability, a correlation between two outputs can be estimated and taken into account. For example, if the Validator includes two outputs that each produce a false positive 1 in 10, the Validator may produce a total of 1 in 100 false determinations; its error probability is then 1/100. If the outputs or the signals provided at them are subject to a correlation greater than zero, the overall error rate of the validator can be higher and an error can be expected in 2 out of 100 cases, for example.

According to a further aspect of the present invention, a positioning system comprises N mutually independent information sources for information indicative of a geographical position; a validator associated with the information sources described herein; and means for determining a position based on the information.

The positioning system can be used in particular to determine a geographic position of a vehicle, preferably a motor vehicle. According to another aspect of the invention, a vehicle includes a positioning system as described herein. The information sources can be included in the vehicle. Example information sources include a GNSS receiver, a radar sensor, a LiDAR sensor, a camera, or an ultrasonic sensor. Information indicative of the position can be provided by processing on the basis of a scanning of an environment of the vehicle by means of a sensor. For example, based on a camera image of the surroundings, the position of the vehicle can be determined with respect to one or more landmarks that are shown on the camera image and whose geographic positions can be known.

According to yet another aspect of the present invention, a method for providing a validator for N mutually independent pieces of information indicative of a geographic position comprises the steps of connecting each of N inputs of an ML system to an information source for one of the N pieces of information; and training the ML system to provide a predetermined signal at one of M outputs if and only if one of the information at the inputs matches a predetermined position hypothesis worse than a predetermined degree, the ML system being simultaneously trained to to avoid incorrect missing signal on two or more outputs.

The ML system is set up for machine learning and can be implemented, for example, in the form of an artificial neural network. The trained ML system may correspond to a validator described herein.

Due to the additional learning objective of avoiding incorrect signals - provided or absent - simultaneously at several outputs, the ML system can develop in such a way that a signal determined at one output is predicted or estimated at another output. An output can focus on validating the information at one of the inputs and the determination of signals on other outputs related to that input can be suppressed. Different assignments of inputs to outputs can develop for different operating states of the validator. A number of inputs to the ML system usually corresponds to a number of available information sources. A number of outputs of the ML system is not necessarily dependent on a number of inputs and is preferably selected in such a way that a determination accuracy of the ML system as a validator is maximized.

Training an ML system to become a validator can be done iteratively. For this purpose, initially only a single output of the ML system can be considered during training. Whenever the outputs considered show sufficient recognition performance, one of the other outputs can also be considered. The training can be continued considering the one exit more. Training considering a predetermined number of outputs usually requires iterations, each Include information at the various inputs related to a common point in time or to a common position.

The training can include a cost function, each of which has a predetermined penalty for an incorrectly missing signal at one of the considered outputs; an incorrectly provided signal at one of the considered outputs; and an incorrect missing signal at two or more considered outputs. The penalty determined using the cost function may be increased if two or more of the above conditions are met. Predetermined weights can be assigned to the outcomes under consideration, with a penalty being dependent on the weights of outcomes that are assigned to a fulfilled condition for a penalty.

In addition to the cost function, a reward function can be used to promote desired behavior of the ML system. The learning can be done such that the ML system is adjusted to minimize a value of the cost function and/or to maximize a value of the reward function.

A reward function can each have a predetermined bonus for a correct signal at an output; and correct signals at multiple outputs. A correct signal can refer to a correctly missing signal or a correctly provided signal. A bonus determined using the rewards function can also be increased if several conditions are met. In particular, the bonus can be increased depending on a number of correct signals.

The training may include a reward function that reduces a bonus for a correct signal at an output if a correct-no-signal rate for that output exceeds a predetermined threshold. This can prevent one output from using an excessively large safety margin to achieve an extremely low false positive rate, which can result in a degraded determination rate of another output.

• 1 ein System;
• 2 ein Ablaufdiagramm eines Verfahrens;
• 3 ein Positioniersystem zur Bestimmung einer Position; und
• 4 ein Ablaufdiagramm eines weiteren Verfahrens;

illustriert.The invention will now be described in more detail with reference to the accompanying drawings, in which:

• 1 a system;
• 2 a flowchart of a method;
• 3 a positioning system for determining a position; and
• 4 a flowchart of a further method;

illustrated.

1 FIG. 1 shows a system 100 that includes a vehicle 105, in particular a motor vehicle such as a passenger car or a motorcycle, and a device 110 that is mounted on board the vehicle 105. FIG. The vehicle 105 can be controlled depending on a specific geographical position, for example in the longitudinal and/or lateral direction. Device 110 includes at least one information source 115 for providing information that indicates a position and/or orientation of vehicle 105 . As will be explained in more detail later, a plurality of information sources 115 are preferably provided. An information source 115 can include a sensor that can be set up in particular to scan an environment of vehicle 105 . Multiple information sources 115 can also use a common sensor. In particular, the sensor can include a camera, a radar sensor, a LiDAR sensor or an ultrasonic sensor. A receiver for signals from a preferably satellite-based navigation system can also serve as a sensor. Known such systems include GALILEO, GPS and GLONASS. Determinations can each be improved on the basis of differential measurements (DGPS: differential GPS).

It is further preferred that each information source 115 uses only one sensor to ensure statistical independence of the information provided. A processing device 120 is provided for processing information provided by the information source 115 . Furthermore, a reference 125 can be provided, which provides information that can be assigned to several or all information sources 115 . For example, the reference 125 can include a map memory for providing map data or an odometer for providing odometry data. The processing device 120 can be connected to an interface 130, which is preferably configured to provide a specific position and/or orientation.

It is proposed that processing device 120 be set up to determine an individual position of vehicle 105 on the basis of information provided by various information sources 115 . To do this, it can use information from reference 125. For example, the information source can include a camera that captures landmarks in the area surrounding vehicle 105 . position Information about the landmarks can be included in the geographical environment information of the reference 125, so that the individual position can be determined by fitting the observations to the known information. This process is also called “matching” and an executing element is also called “matcher” or adaptor. A matcher is typically provided specific to a predetermined information source 115 . Different, specialized adapters can be provided for different driving situations.

The processing device 120 can also provide a position hypothesis, on the basis of which a subsequent determination can work, for example. The position hypothesis can be determined, for example, on the basis of a previously determined position and a movement of the vehicle 105 that has taken place since then. Alternatively, the determined position can be used as a position hypothesis. During the matching, environmental information of the reference 125 can be taken into account in a predetermined area around the position hypothesis.

Based on the individual positions, the processing device 120 can determine a position of the vehicle 105 . In addition, the processing device 120 is preferably set up to determine for the individual information sources 115 how well the information provided matches the specific position or the position hypothesis. For this purpose, the respective information and the individual position determined from it can be viewed. The process is also called validation and an executing component validator. A matcher and a validator can be integrated with each other.

If the information does not match the specific position as well as a predetermined measure, the processing device 120 can provide a signal that can be provided via the interface 130 . The signal can be output if the information from at least one of a plurality of information sources 115 matches the specific position worse than predetermined. The signal can indicate that no position can be determined or that the position could not be determined with sufficient certainty. A control dependent on the determined position, for example for a movement of the vehicle 105, can then be deactivated and a driver can be requested to take over control of the vehicle 105.

2 shows a flow chart of a method 200, which can be executed in particular on board a vehicle 105 and preferably by means of a device 110. The method 200 comprises a number of function blocks 205 which can be executed concurrently with one another. The same or different elements of the device 110 can be used for this purpose. Although any number of function blocks 205 can be formed, their number in a typical vehicle 105 is generally around 2-4.

At a function block 205 , an information source 115 provides information indicative of a vehicle 105 position, orientation, or pose. A pose generally includes a position and an orientation. For example, in three-dimensional space, a pose may include six statements, three of which are translational and three of which are rotational with respect to a Cartesian coordinate system. Although the invention could also be implemented on the basis of a position or an orientation instead of a pose, poses are referred to in the following.

In a step 210, based on the information provided by the information source 115, a single pose is determined. In the present case, an “individual” pose is used when it is determined on the basis of just one information source 115 . The pose of the vehicle 105 is determined based on a plurality of "individual" poses. Step 210 may be performed in any manner and may include, for example, averaging, an adaptive filter, a physical model of the vehicle 105, or a trained neural network. The determination can be made based on still other information provided by the reference 125 . Step 210 can also be feedback and process a previously self-provided single pose.

A basic check can be made for a particular pose. For example, it can be determined whether vehicle 105 is located on a route on which a driving function to be controlled is actually offered. For example, a prerequisite for following a lane can be that the route includes a freeway or an expressway. Such a check can also be carried out using a separate function block. If a required prerequisite is not met, the method 200 may terminate.

In a step 215 the information from the information source 115 used in step 210 is evaluated by checking how well it matches a pose of the vehicle 105 . This determination may also be called validating, and an element performing step 215 may be called a validator. The pose of the vehicle 105 is determined based on certain individual poses, as detailed below. A distance between the individual pose and the pose of the vehicle 105 can be determined for the evaluation. The distance may include a Euclidean distance for each included location and an angle between each included orientations. The distance and angle can be used to form a combined index to facilitate later comparison to a predetermined measure. The predetermined amount may be less than or equal to a maximum tolerable error in the particular pose. An individual measure can be predetermined for each function block 105 . If the distance or the code number exceeds the predetermined amount, a signal, in particular a warning signal, can be provided to a step 225 .

The predetermined measure can be determined as a function of a driving condition and/or a driving function of vehicle 105 . The driving function can in particular include a driving assistant that is set up to relieve or support a driver of the vehicle 105 . The driving assistant can carry out a longitudinal control and/or a lateral control of the vehicle 105 or intervene in one of them if necessary. In one specific embodiment, the driving assistant is set up for automatically or autonomously guiding the vehicle 105 .

Should the individual pose match the specific pose of vehicle 105 worse than predetermined, for example if the distance exceeds the predetermined amount, then a signal can be provided which can be evaluated in step 220 . In addition, the pose can be determined in step 210 as a function of the quality determined, the distance determined, or the signal determined. For example, in step 210 an adaptive filter may be calculated, which may be reinitialized if the individual pose provided by the filter deviates from the pose of the vehicle 105 more than predetermined.

In a step 220 the pose of the vehicle 105 is determined. For this purpose, the individual poses determined in the function blocks 105, in particular the steps 210, are preferably combined with one another. The combination can take into account all determined individual poses in the same way or individually. For example, individual poses determined on the basis of different processing of information from the same sensor can be taken into account. Individual poses, or the data on which they are based, which were determined to be untrustworthy by one of steps 215 can be discarded when determining the pose of vehicle 105 . The determined pose of the vehicle 105 can be provided to the outside.

In a step 225, validation signals from the function blocks 205 can be evaluated. At interface 130 a signal indicative of a potentially untrustworthy particular pose of vehicle 105 may be determined and provided based on one or more signals from steps 215 of functional blocks 105 .

In one embodiment, such a signal is already output at the interface 130 if at least one of the specific individual poses deviates from the specific pose by more than the associated predetermined amount. This can be equivalent to the presence of at least one warning signal in one of steps 215 .

In another embodiment, in a step 215, not only can a warning signal be provided or not provided, but also no statement can be made about a specific pose. For example, a LiDAR sensor in the vicinity of the vehicle can scan another vehicle that does not correlate with any object noted in a map memory. In this case, a single pose determined based on other LiDAR sensor scans cannot be confirmed or disputed. In such a case, an assessment of the specific individual pose in step 215 can be omitted. As a result, at step 225, the warning signal at interface 130 cannot be determined based on the missing signal. However, the warning signal may still be given if another validator 215 determines doubts about the trustworthiness of an associated individual pose, or too few validators 215 evaluate the pose. If in none of the steps 215 of the function blocks 205 has a statement been made regarding the trustworthiness of the data under consideration, the warning signal can also be output in step 225 .

3 shows another system 100 for determining a geographic position, in particular of a vehicle. In this embodiment, a validation function of the functional blocks 205 of the embodiment of FIG 2 provided by an ML system (machine learning system) 305 . The ML system 110 can therefore also be spoken of as a validator 110 or as a combination of several validators 110 .

The ML system 110 includes N inputs 310 and M outputs 315. The ML system 110 can be embodied in particular as an artificial neural network. The inputs 310 can Nodes of an input layer and the outputs 315 correspond to nodes of an output layer of the network. The network may include one or more intermediate layers. In general, the number N of inputs 310 corresponds to a number of available information sources 115. In this case, one information source 115 is preferably connected to exactly one input 310 in each case. The number M of the outputs 315 can deviate from the number N.

A position hypothesis, which can be provided at the interface 130, can be determined by means of the processing device 120 on the basis of the information provided. Samples and map data of the reference 135 can be used to determine the position hypothesis. Other sources of information can also be used. For example, the position hypothesis may be determined based on a previously determined position and movement of the vehicle 105 .

The ML system 305 is set up to determine for the information from the information sources 115 in each case whether or not they match a specific position hypothesis sufficiently well. To this end, as described above, individual positions, orientations or poses can be determined on the basis of information from an information source 115 and a deviation of the result from another determination can be determined. If information does not match the position hypothesis, the ML system 305 can provide a signal at one of its outputs 315 indicating this. Otherwise no such signal can be provided. A fixed association between an input 310, at which information from an information source 115 associated with it is provided, and an output 320 is not required. Although such an association may exist, it may change depending on the information being processed.

An optional combination device 320 is set up to determine an integrity of a position that can be determined on the basis of the information. In particular, the combination device 320 can determine a signal that indicates whether the information from the information sources 115 with regard to the determination of a position, orientation or pose matches better than predetermined. The signal can be provided at an interface 325 . The position hypothesis determined by means of the processing device 120 can only be provided as a position at the interface 130 if the signal from the combination device 320 indicates that the information fits together sufficiently well.

It is proposed to train the ML system 305 to avoid an incorrect missing signal on two or more outputs 315 (false positive). For this purpose, a cost function can be used when training the ML system 305, which assigns increased costs to multiple false positives.

4 FIG. 4 shows a flow diagram of a method 400 for training an ML system 305 to be a validator for information indicative of a geographic location. The method 400 represents a preferred procedure for training the ML system 305 for use in the positioning system 100. The method 400 implements iterative learning, in which initially only one output 315 of the ML system 305 is considered and gradually other outputs 315 may be added until the entire ML system 305 is sufficiently trained.

In a step 405, a run variable i is set to one. The control variable i indicates how many of the outputs 315 are currently being considered when the ML system 110 is learning. For example, if i is one, only a first output 315 is considered; if it is two, the first two outputs 315 are considered.

In a step 405, the outputs 315 one to i are considered, i.e. only the first output 315 in a first run of the method 400. Under this condition, the ML system 305 is trained in a step 415. Real, determined in real time , or also recorded sample values from information sources 115 on board a vehicle can be used during a test drive. An actual position of the vehicle 105 in relation to the information is usually known. The training may include determining a position hypothesis and determining the integrity of the information at a time. Based on the known actual position, it can be understood whether an assessment determined by the ML system 305 as to whether information at one of the inputs 310 matches the position hypothesis sufficiently well or not is correct.

A reward function and/or a cost function may be provided for training. The reward function may determine a bonus if a correct signal at an output 315 is determined. Another bonus can be determined when correct signals on multiple outputs 315 are determined. A bonus for a correct signal at an output may be reduced if a correct missing signal rate for that output exceeds a predetermined threshold.

The cost function can determine a penalty if an incorrectly missing signal is determined at one of the considered outputs, an incorrectly provided signal is determined at one of the considered outputs; or an incorrect missing signal is determined at two or more considered outputs. Predetermined weights can be assigned to the outputs 315, it being possible for a penalty, which relates to one of the outputs 315, to be determined as a function of the weight.

The ML system 305 can be adjusted to maximize a bonus determined by the reward function and/or to minimize a penalty determined by the cost function. In the case of an artificial neural network, for example, this process can be carried out using backpropagation. Weights of individual nodes can be changed so that the described effect occurs.

In a step 420 it can be determined whether a recognition performance of the currently considered outputs 315 is sufficiently good with respect to a predetermined criterion. If not, the method 400 may return to step 410 and run through again. Otherwise, if the recognition performance is found to be sufficient, the control variable i can be incremented in a step 425. Then, in a step 430, it can be determined whether i exceeds the number M of exits 315 present. If this is not the case, the method 400 can return to step 410 and be trained further taking into account a further outcome 315 . Otherwise, all outputs 315 are already taken into account during training, so that no more outputs 315 can be added. The method 400 can end in this case or the ML system 305 can be further trained considering all outputs 315 .

The training includes a target that causes multiple simultaneous false missings to be specifically avoided. This organizes the ML system so that an output 315 focuses on the information at one of the inputs 310 . The trained ML system 110 can behave like multiple individual validators, each associated with one of the inputs 310 . These validators can be essentially statistically independent of each other; a remaining correlation between two validators can be estimated, for example based on observations of the ML system during the integrity determination.

A probability that information provided at the inputs 310 of the ML system 305 does not match sufficiently well, although no signal is provided at an output 315 indicative of ill-matched information, can be incorrect as a product of the probabilities because of the independence of the individual validators Provisions of the individual validators are determined. If this is below a predetermined threshold value, then the ML system 305 can be sufficiently trained.

A recognition performance of each such validator can be determined individually. It is true that a poorly performing validator, which provides, for example, one incorrect determination per 20 pieces of information examined, can be determined quickly. A well-working validator, on the other hand, which, for example, only provides one wrong determination per 1000000 pieces of information examined, requires a significantly higher number of pieces of information in order to be able to assess its quality.

An unsatisfactory validator can be merged with another by merging the corresponding outputs 315 . A number of outputs 315 of the ML system 305 can be reduced as a result. Conversely, the number of outputs 315 can also be increased by using nodes of the ML system 305 to set up a further validator which, if possible, does not correlate with existing validators. If the ML system 305 is then further trained, correlations between the validators should be further reduced.

Additionally, the ML system 305 can be trained or selected to maximize availability. It may be required that the availability exceeds a predetermined threshold value, so that, for example, the integrity of the information cannot be determined in only one out of 100,000 cases. The availability can be 99.999% in this example.

Reference List

100

system

105

vehicle

110

contraption

115

source of information

120

processing facility

125

reference

130

interface

200

procedure

205

function block

210

determine pose

215

Evaluate and validate the signal

220

determine pose

225

provide signal

305

ML system

310

Entry

315

Exit

320

combination device

325

interface

400

procedure

405

i := 1

410

Consider outputs 1 to i

415

Work out

420

Recognition performance of observed outputs high enough?

425

i = i + 1

430

i ≤ M ?

QUOTES INCLUDED IN DESCRIPTION

This list of documents cited 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

• DE 102019133316 A1 [0005]

Claims (10)

Validator (305) for N independent pieces of information indicative of a geographic position, the validator (305) having N inputs (310) for the information and M outputs (315) and being trained using machine learning, exactly then at one the outputs (315) to provide a predetermined signal when any of the information matches a predetermined position hypothesis less than a predetermined degree; wherein the validator (305) is further trained to avoid an incorrect missing signal at two or more outputs (315). Validator (305) after claim 1 , further comprising combining means (320) connected to the M outputs (315) for determining an integrity of a position determinable on the basis of the information. A positioning system (100) comprising N independent information sources (115) for information indicative of a geographical position; a validator (305) connected to the information sources (115) according to any one of the preceding claims and means (120) for determining a position based on the information. Vehicle comprising N mutually independent information sources (115) for information indicative of a geographical position of the vehicle, and a positioning system (100). claim 3 . Method (400) for providing a validator (305) for N mutually independent information indicative of a geographic position, the method (400) comprising the following steps: - connecting in each case one of N inputs (310) of an ML system (305) to an information source (115) for one of the N pieces of information; - training (415) the ML system (305) to provide a predetermined signal at one of M outputs (315) precisely when one of the items of information at the inputs (310) matches a predetermined position hypothesis worse than a predetermined degree; - wherein the ML system (305) is simultaneously trained to avoid an incorrect missing signal at two or more outputs (315). Method (400) according to claim 5 , wherein initially only one output (315) is considered during training; and in each case one of the further outputs (315) is additionally considered if the previously considered outputs (315) show sufficient recognition performance. Method (400) according to claim 5 or 6 , wherein the training comprises a cost function each having a predetermined penalty for - an incorrect missing signal at one of the considered outputs (315); - an incorrectly provided signal at one of the considered outputs (315); and - an incorrect missing signal at two or more considered outputs (315). Method (400) according to claim 7 , wherein predetermined weights are assigned to the outputs (315) under consideration; and a penalty is dependent on weights of outputs (315) associated with a met condition for a penalty. Method (400) according to any one of Claims 5 until 8th , wherein the training comprises a reward function each having a predetermined bonus for - a correct signal at an output (315); and - correct signals at multiple outputs (315). Method (400) according to any one of Claims 5 until 9 wherein the training comprises a reward function that reduces a bonus for a correct signal at an output (315) if a correct-no-signal rate for that output (315) exceeds a predetermined threshold.

Images