DE102020212306A1 - Procedure for early tunnel detection - Google Patents

Abstract

Method for early tunnel detection using a radar sensor (6), in which the radar sensor (6) emits a radar signal, the radar signal is reflected by objects and radar detections are recorded on the basis of the reflections, the radar sensor (6) being prepared in such a way that it distance information and altitude information, at least one region of interest is determined (11) within which the radar detections are used for further processing, based on which radar detections a first and a second histogram (12a, 12b) are generated, the first histogram ( 12a) includes the density of radar detections within the region of interest, the second histogram (12b) includes the height information of the radar detections within the region of interest, and early tunnel detection is based on the distance information and a statistical evaluation of the histograms (12a, 12b).

Description

The present invention relates to a method, in particular a computer-implemented method, for early tunnel detection on a route, a system which is designed to carry out early tunnel detection using the method according to the invention, a computer program with program code for carrying out the method according to the invention and a computer-readable storage medium which the computer for Execution of the method causes.

Technological background

Modern means of transportation such as B. motor vehicles or motorcycles are increasingly equipped with driver assistance systems, which use sensor systems to detect the environment, recognize the traffic situation and support the driver, z. B. by a braking or steering intervention or by the output of a visual or acoustic warning. Radar sensors, lidar sensors, camera sensors or the like are regularly used as sensor systems for detecting the surroundings. From the sensor data determined by the sensors, conclusions can then be drawn about the environment, e.g. B. an object and / or environment classification can be made or an environment model can be created. Furthermore, the detection of the environment is almost indispensable in the field of (partially) autonomous driving. By processing the sensor and environment data then assistance functions can be performed, through the z. B. Accidents with other road users can be avoided or complicated driving maneuvers can be made easier by supporting or even completely taking over the driving task or vehicle guidance (partly / fully automated). For example, the vehicle z. B. by an emergency brake assistant (EBA, Emergency Brake Assist) an autonomous emergency braking (AEB, Automatic Emergency Brake), by an Adaptive Cruise Control Assistant (ACC) following driving and cruise control or by an active lane departure warning with steering assistance ( LKA - Lane Keeping Assist; LCA - Lane Centering Assist or LKS - Lane Keeping Support).

Environment detection using radar sensors is often used in modern means of transportation. This is based on the emission of bundled electromagnetic waves (radar waves or radar signals) and their reflection, e.g. B. by other road users, obstacles on the road or the edge of the road. The differentiation of obstacles or objects such. B. other road users, traffic signs, road markings, peripheral buildings, crash barriers and the like is of great importance to determine when z. B. a braking or steering intervention is to be initiated.

In an idealized traffic scenario, the probability that the objects relevant for detecting the surroundings will be detected by the radar sensor is very high. However, the functionality of such radar sensors in noisy environments such. tunnel areas, bridges, underground car parks, closed structures, etc., especially since many different detections can generate ghost objects (i.e. objects that are incorrectly recognized due to detections but do not really exist) in noisy environments. By ghost objects it can then z. B. lead to incorrect braking or unnecessarily triggered braking. Furthermore, the regions in front of a tunnel (the early tunnel regions) are usually also disturbed, with the interference intensity increasing as soon as the tunnel entrance is reached. Conventional tunnel detection algorithms only work if the ego vehicle drives far enough into the tunnel. For example, it has been shown that the emergency braking properties (or EBA functionality) are negatively affected in regions before the tunnel, since a sudden transition in this region (“tunnel” to “non-tunnel” or vice versa) can cause critical detection conditions. There is therefore a special interest in continuously and promptly recognizing tunnel entrances in all environmental and traffic scenarios. For example, a tunnel detection or tunnel entry can be detected with a radar sensor.

Printed state of the art

From the DE 10 2018 207 855 A1 a method for operating a motor vehicle is known, in which radar data from a radar sensor directed towards the area in front of the motor vehicle is evaluated in order to detect an upcoming end of the tunnel. The end of the tunnel is determined using an angular resolution in azimuth and elevation of the radar sensor. Due to the angular resolution in the elevation, the radar sensor is able to differentiate whether a reflection comes from above the motor vehicle, e.g. B. from the tunnel ceiling. In the same way, this method can also be used to detect the start of a tunnel ahead in the direction of travel of the motor vehicle.

Object of the present invention

The object of the present invention is now to provide a novel method for tunnel detection which improves the tunnel detection in a simple and cost-effective manner and with which the disadvantages known from the prior art are overcome.

solution of the task

The above object is achieved by the entire teaching of claim 1 and the independent claims. Expedient developments of the invention are claimed in the dependent claims.

For tunnel detection, the method according to the invention uses early tunnel detection or tunnel pre-classification in order to detect the tunnel ahead at a certain distance before entering the tunnel, depending on how far away the tunnel is from the ego vehicle or until it meets the tunnel ( TTC or time-to-collision). The main reason for using such an early tunnel detection is to improve the propulsion functionality or the sensor fusion by allowing different parameterizations based on the tunnel information recorded. This method is particularly suitable for vehicles that detect the surroundings, e.g. B. for object recognition and object classification and for the execution of assistance functions. According to the invention, a radar sensor is provided for detecting the surroundings, which emits a radar signal (i.e. radar beams or an electromagnetic wave) which is reflected by the objects in the vicinity of the radar sensor, so that radar data or radar detections are generated based on the reflected radar signals. These radar detections can then be used to identify the objects in the vicinity of the radar sensor, e.g. B. other road users (cars, trucks, pedestrians, two-wheelers, scooters, animals and the like), roadside buildings, vegetation, crash barriers and the like. Here, distance information is assigned to the detections or objects. Furthermore, the radar sensor can assign height information to the radar detections, i. H. the radar sensor can, for example, determine the elevation of the radar detections or of the objects assigned to the radar detections. Furthermore, at least one region of interest is defined within which the radar detections are used for further processing. A first and a second histogram are generated on the basis of the radar detections. The first histogram includes the density of radar detections within the region of interest (density histogram) and the second histogram includes the altitude information of the radar detections within the region of interest (altitude histogram). The early tunnel detection is then based on a statistical evaluation of these histograms. The method according to the invention creates the advantage that sudden changes in the environment are identified far in front of the tunnel region. As a result, safety-critical functions or assistants can be improved by being able to draw their attention to changes in the environment at an early stage before these changes have a critical effect on the environment detection or the driving of the ego vehicle. Furthermore, early tunnel detection can be used in semi-automated driving to inform or force the driver to take or take control of the vehicle again. For example, the algorithm recognizes that the ego vehicle can no longer be controlled reliably as soon as a tunnel ahead is detected.

The distance information and/or a tunnel probability of the tunnel or of the object that is to be classified as a tunnel is preferably determined using the statistical evaluation of the histograms that have been created. For example, the histogram may include a discrete, randomized approximation of the distribution. This means that well-known statistics can be easily applied to any distribution and interpreted with a sound theoretical background. Advantageously z. B. The maximum (the maximum of the histogram as a statistic would be called "mode") and the average are used for the calculation. Alternatively or additionally, however, more complex statistics, such as e.g. B. the median can be used, which are in a simple, monotonic relationship to the tunnel probability and distance. In a simple way, the position of the maximum value (argmax) could also be used to estimate the distance to the tunnel.

Furthermore, such a statistic could be used as an input (function) of a machine learning (ML) algorithm, the term “machine learning (ML)” or “machine learning” in the sense of the invention describing the use of algorithms or models, to analyze data, learn from it, and then make a determination or prediction about something (not yet visible). Both supervised and unsupervised learning can be used. For the analysis of the data, the labeling strategy or labeling or labeling of the data is of crucial importance. Alternatively or additionally, the histogram as such can also be fed directly into the ML model as an input, depending on the capabilities of the model. Equally, this also applies to more complicated statistics that do not have a simple (monotonic) relationship with tunnel probability and distance, such as B. the standard deviation, moments, kurtosis, skewness and the like. The methodological difference between early tunnel detection and distance estimation is that early tunnel detection e.g. B. is to be regarded as a binary, classified task for which standard ML models already known can be used in a simple manner and the distance estimation is a so-called regression task, since the distance is non-binary and is present as a real number, in which the other ML models must be applied. The input functions can therefore be similar for both models, but differ depending on the input. For example, linear regression can be used for distance estimation and logistic regression for binary classification.

The statistical evaluation can expediently be carried out using the result of the statistical mean value of the first histogram and the maximum height of the second histogram.

Furthermore, the radar detections of the radar sensor can be filtered using filtering before further processing. The filtering can z. B. done as a mathematical sub-algorithm. In particular, the filtering can be used when detecting stationary detections or objects.

The radar detection is influenced by the reflecting objects in the vicinity of the radar sensor or the ego vehicle and their material properties (reflectivity). Some radar detections have multiple radar signal propagation paths and consequently multiple hypotheses (targets). Such detections are marked as multi-targets or multiple targets, in particular of an object. Advantageously, the radar detections for early tunnel detection can also include such multi-targets in order to further improve the early tunnel detection.

The region of interest is preferably defined using at least one driving parameter, in particular using the speed and/or the acceleration of the ego vehicle.

The method preferably includes a first sub-algorithm for freeway scenarios and a second sub-algorithm for city scenarios, with the sub-algorithms being executed one after the other or in parallel. In addition, further sub-algorithms can be provided in order to further increase the reliability of the determination.

Furthermore, in the first sub-algorithm, the region of interest can be affected in the lateral direction by detected static objects, i. H. for example crash barriers or roadside buildings, since these can usually be reliably and safely determined in freeway scenarios using the available sensors (radar, lidar, camera and the like).

In addition, in the second sub-algorithm for urban scenarios, the curve factor can be used to determine the tunnel probability, e.g. B. to map frequent steering and / or maneuvering. The curve factor describes the curvature information of the route (e.g. determined via inclination angle measurements, steering angle, navigation data and the like), which can be used instead of a street estimate, since street estimates are not always reliable in urban scenarios. To reduce the impact of sudden turns and U-turns, e.g. e.g. when changing lanes, the curvature can be extracted or filtered out (i.e. the curve factor can be used to determine the region of interest, with the curvature information being filtered to calculate the corresponding maneuvers). In addition, the reversal factor, which corresponds to the curvature angle of the travel path of the ego vehicle, can be used to further improve the reliability of the system.

In a preferred manner, the results of the sub-algorithms can then be merged, so that the tunnel probabilities or early tunnel detections thus determined are compared with one another or supplemented. As a result, a kind of function check can be carried out by e.g. B. stored in a memory of the control unit comparison pattern or calculation modules can be used to compare the results of the individual sub-algorithms (plausibility check).

According to a special embodiment of the method, an environment classification or environment model is created, which is used to define the at least one region of interest and/or to select the respective sub-algorithm. The objects detected by the radar sensor can preferably be used for environmental classification. Furthermore, other parameters can also be used to classify the environment, e.g. B. a traffic sign and / or road user recognition by means of a camera, 3D data from appropriate sensors (z. B. lidar / high-flash lidar, camera / stereo camera or the like) and / or navigation / GPS data.

In a practical way, based on the radar detections, a reflection pattern can be recorded, which is used for the classification of the surroundings. For example, environmental features such as B. many slow-moving objects, increased traffic, frequent Making turns and U-turns, lots of pedestrians and the like, shooting at an urban environment. In the same way, fast-moving road users (high speeds, e.g. more than 60 km/h, in particular more than 80 km/h or more than 100 km/h) and comparatively small changes in the direction of movement could indicate a motorway environment. Exemplary reflection patterns that are suitable for environment classification can e.g. B. be stored in a memory of the control device, in order to then carry out the environment classification based on a comparison of the detected reflections with the stored reflection patterns. In addition, the reflection patterns can also be learned or trained using ML algorithms in order to increase the number of scenarios to be recognized.

The height information can expediently be derived through a filter before being assigned to the respective radar detection. A filter is advantageously used, in particular a 1D Gaussian filter (Gaussian filter), in order to describe the empirical distribution of the height information in the real tunnel, so that the height information can be derived before it is assigned to the respective radar detection. In doing so, e.g. B. radar detections or objects to these detections with a certain height above the road, z. B. of more than 10 m and less than 2 m, preferably more than 8 m and less than 3 m and particularly preferably more than 7 m and less than 4 m are filtered out or discarded. This allows object heights to be selected that would also correspond to a real tunnel entrance. Furthermore, the upper limit can also be omitted since it can be limited by the elevation determination limits of the radar sensor.

The present invention also claims a system, in particular a data processing system, sensor system or a system comprising a sensor device, with a radar sensor and a computer or a control device, which is set up to carry out early tunnel detection using the method according to the invention.

The present invention also includes a computer program with program code for carrying out the method according to the invention when the computer program is executed in a computer or another programmable computer known from the prior art (e.g. a computer device comprising a processor, microcontroller or the like).

In addition, the present invention comprises a computer-readable storage medium comprising instructions which cause the computer on which they are executed to carry out a method according to at least one of the preceding claims.

Surprisingly, it has also been shown that the present invention can also be applied to other types of sensors, such as e.g. B. camera, lidar, ultrasonic sensors or sensor combinations thereof is applicable if suitable input data is used and the training is tailored accordingly.

Furthermore, the method according to the invention can be a purely computer-implemented method, with the term “computer-implemented method” in the sense of the invention describing a flow plan or procedure that is implemented or carried out using a computer. The calculator, such as B. a computer, a computer network, a control unit (z. B. ECU or Electronic Control Unit or ADCU or Assisted & Automated Driving Control Unit) or a computing device therein or any other programmable device known from the prior art, can process the corresponding data using programmable calculation rules. With regard to the process, the essential properties can be e.g. B. be effected by a new program, new programs, an algorithm or the like.

Description of the invention based on exemplary embodiments

• 1 eine stark vereinfachte schematische Darstellung eines Ego-Fahrzeuges mit einer Tunnelfrüherkennung anhand des erfindungsgemäßen Verfahrens:
• 2 eine vereinfachte schematische Darstellung einer Ausgestaltung eines erfindungsgemäßen Verfahrensablaufs, sowie
• 3 eine vereinfachte schematische Darstellung im 1 D-Diagramm einer Ausgestaltung eines Gauß-Filters.

In the following, the invention is explained in more detail by means of expedient exemplary embodiments. Show it:

• 1 a greatly simplified schematic representation of an ego vehicle with early tunnel detection using the method according to the invention:
• 2 a simplified schematic representation of an embodiment of a method sequence according to the invention, and
• 3 a simplified schematic representation in the 1D diagram of an embodiment of a Gaussian filter.

Reference number 1 in 1 refers to an ego vehicle with various actuators (steering 3, motor 4, brake 5), which has a control device 2 (ECU, Electronic Control Unit or ADCU, Assisted and Automated Driving Control Unit) through which a (partially) automated control of the ego vehicle 1 can take place, e.g. B. in that the control device 2 can access the actuators of the ego vehicle 1 . The ego vehicle 1 also has sensors for detecting the surroundings: a long-range radar sensor 6, several short-range radars sensors 7a-7d, a lidar sensor 8 and a front camera 9), the sensor data for environment and object detection are used, so that various assistance functions, such. B. Emergency brake assistant (EBA, Electronic Brake Assist), distance control (ACC, Automatic Cruise Control), lane keeping control or a lane departure warning system (LKA, Lane Keep Assist) or the like can be realized. The radar sensor 6 is a long-range radar sensor that ensures detection of the surroundings even at higher speeds (e.g. through a wide range of >200 m) and which has an elevation measurement or height angle measurement in order to obtain height information on the detected objects to capture. For example, the height information using a method EP 2 294 446 B1 or another method known from the prior art. Furthermore, distance information, e.g. B. based on the design of the radar sensor according to an FMCW (Frequency Modulated Continuous Wave) method, using Doppler radar or the like. The assistance functions are executed via the control device 2 or the algorithm stored there.

The ego vehicle 1 includes early tunnel detection according to the invention, which is implemented in particular using the data from the radar sensor 6 . The method sequence according to the invention can be carried out as a purely computer-implemented method on the control device 2, which thus serves as a computer. The goal of early tunnel detection is to detect the tunnel at a certain distance before the ego vehicle 1 enters the real tunnel, in order to avoid z. B. to support assistance functions. The specified distance can be based on a time to enter the tunnel or "Time to Collision" (TTC). An embodiment of a process flow according to the invention is in 2 shown.

The radar detections or radar data of the radar sensor 6 are first filtered before further processing (filtering 10), with a region of interest or ROI (region of interest) being defined (definition 11 for city scenario/motorway scenario), within which the radar detections are used for further processing . Furthermore, different regions of interest, e.g. B. differentiated based on the ego speed. The pattern information or radar information is recorded on two 1-D histograms 12a, 12b for early tunnel detection or early tunnel detection in city and freeway scenarios. The first histogram 12a maps the density of the detections in a specific region (density histogram), while the second histogram 12b forms the combined height information of the detections (height histogram). The height information is z. B. derived by a filter, the z. B. is similar to a 1D Gaussian filter, which weights the tunnel height more. This allows statistical information to be extracted from the histograms 12a, 12b. For example, the result of the statistical mean value of the detection histogram and the maximum height of the histogram can be used for early tunnel detection or tunnel pre-classification. The Gaussian filter can be designed as a low-pass filter that is used to reduce noise. As a spatial filter, it can be used to enhance the data or regions. An exemplary embodiment of a Gaussian filter in the 1D diagram is given in 3 shown. Furthermore, further improvements can be made especially for the early detection (ie the detection of the early tunnel areas - just before entering the tunnel) of motorway and city tunnels.

The radar reflections of the radar sensor 6 can be influenced, for example, by the environment, the reflecting objects in the environment and the material properties (reflectivity) of the objects, so that the pattern of the radar reflection varies as soon as there are changes in the environment. The recognition of these different patterns and the extraction of these is used for the recognition of disturbed environmental areas (recognition and extraction 13). Due to the fact that the environment outside the tunnel is different in city and highway regions, the algorithm can e.g. B. on the basis of differently weighted and recorded parameters in two parallel configured algorithms (sub-algorithms). As a result, traffic scenarios for city and freeway tunnels can be handled better by the algorithm (sub-algorithm 14 for freeway scenarios or sub-algorithm 15 for city scenarios). The respective sub-algorithm 14, 15 can then be selected on the basis of the recognition and extraction 13. In a practical manner, further sub-algorithms can also be provided in order to determine further special traffic scenarios (e.g. tunnels on country roads or the like) in each case via a separately running sub-algorithm. The more individual scenarios are covered by a sub-algorithm, the greater the operational and road safety. However, since this also increases the computing and programming effort, a compromise must be made between operational and road safety as well as computing and programming effort, which can ultimately determine the number of sub-algorithms.

In well-known freeway scenarios in which a freeway tunnel is to be detected, detections in connection with curve scenarios, crash barriers, noise protection walls and the like are among the major challenges for the radar system. In particular, the detection of static objects also plays a special role here, static objects being stationary objects within the meaning of the invention, e.g. B. crash barriers, buildings, stationary vehicles and the like, which allow the detection of lane boundaries for the ego vehicle 1 in a simple manner. In order to avoid such influences, different regions of interest can be used for different traffic scenarios. For highways z. B. an ROI can be used that was modified based on road boundary information. The lateral boundaries are defined by road edges, while the longitudinal area is fixed (e.g. predefined in meters). The features for classifying a tunnel are extracted from the detections within the defined region of interest.

In urban traffic scenarios where z. B. a city tunnel is to be recognized, the ego vehicle 1 usually drives at a lower speed, so that the region of interest or ROI is preferably defined in longitudinal and lateral distances. Frequent maneuvering or different driving maneuvers, which are usually carried out in urban areas, often lead to "false positive" tunnel detections. In order to avoid these false-positive tunnel detections in the city, the arc radius of the ego vehicle 1 is first filtered in order to determine sudden maneuvering and to calculate the weight for the filtered arc radius. The feature extracted from the respective histogram is weighted on the basis of the curve radius. In addition, several goals can be determined, e.g. B. be extracted by means of a trapezoidal detection region 16 or ROI. The respective additional functions can then z. B. be combined using a linear regression model 17.

The sub-algorithms can then be combined, for example as part of a subsequent data fusion 18. Such a data fusion can take place taking into account the weighted speeds in order to combine the city and freeway scenarios. Accordingly, the respective sub-algorithm can be selected based on the current ego speed of the ego vehicle 1 for early detection of city and freeway tunnels.

The method according to the invention can thus be used to identify and extract important statistical information from the relevant radar data in order to distinguish between tunnel areas and non-tunnel areas at an early stage. The algorithm can be used to identify regions that are less affected by adjacent structures (e.g. tunnels). Furthermore, the algorithm can also be extended in such a way that e.g. B. closed structures such as multi-storey car parks, underpasses, bridges and the like can be recognized and classified.

Reference List

1

ego vehicle

2

control device

3

steering

4

engine

5

brake

6

Long range radar sensor

7a-7d

Short-range radar sensor

8th

lidar sensor

9

front camera

10

filtering

11

determination

12a

density histogram

12b

elevation histogram

13

Detection and Extraction

14

partial algorithm

15

partial algorithm

16

trapezoidal region of interest

17

linear regression model

18

data fusion

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 102018207855 A1 [0005]
• EP 2294446 B1 [0028]

Claims (16)

Method for early tunnel detection using a radar sensor (6), in which the radar sensor (6) emits a radar signal, the radar signal is reflected by objects and radar detections are detected based on the reflections, wherein the radar sensor (6) is set up in such a way that it can assign distance information and height information to the radar detections, at least one region of interest within which the radar detections are used for further processing is determined (11), a first and a second histogram (12a, 12b) are generated on the basis of the radar detections, wherein the first histogram (12a) includes the density of radar detections within the region of interest, the second histogram (12b) comprises the height information of the radar detections within the region of interest, and the early detection of the tunnel is based on the distance information and a statistical evaluation of the histograms (12a, 12b). procedure after claim 1 , characterized in that a tunnel probability and distance information is determined using the statistical evaluation. procedure after claim 1 or 2 , characterized in that the statistical evaluation is based on the result of the statistical mean value of the first histogram (12a) and the maximum height of the second histogram (12b). Method according to at least one of the preceding claims, characterized in that the radar detections of the radar sensor (6) are filtered using a filter (10) before further processing. Method according to at least one of the preceding claims, characterized in that the radar detections for early tunnel detection also include multi-targets. Method according to at least one of the preceding claims, characterized in that the region of interest is determined (11) using at least one driving parameter, in particular the speed and/or the acceleration of the ego vehicle (1). Method according to at least one of the preceding claims, characterized in that the method comprises a first sub-algorithm (14) for freeway scenarios and a second sub-algorithm (15) for city scenarios, the sub-algorithms (14, 15) being executed one after the other or in parallel. procedure after claim 7 , characterized in that in the first sub-algorithm (14) the region of interest is defined in the lateral direction by detected static objects, in particular crash barriers or roadside buildings. Procedure according to one of Claims 7 or 8th , characterized in that in the second sub-algorithm (15) the curve factor is used to determine the tunnel probability, with curvature information being determined and in particular filtered so that it can be used to define the region of interest. Procedure according to one of Claims 7 - 9 , characterized in that a fusion of the results of the sub-algorithms (14, 15) takes place. Method according to at least one of the preceding claims, characterized in that an environment classification takes place which is used to determine (11) the at least one region of interest and/or to select the respective sub-algorithm (14, 15). procedure after claim 11 , characterized in that based on the radar detections, a reflection pattern is detected, which is used to classify the environment. Method according to at least one of the preceding claims, characterized in that a filter is used to describe the empirical distribution of the height information in the real tunnel, so that the height information can be derived before it is assigned to the respective radar detection. System comprising a radar sensor (6) and a control device (2), the system being set up to carry out early tunnel detection using a method according to at least one of the preceding claims. Computer program with program code for carrying out a method according to at least one of Claims 1 - 13 , when the computer program is executed in a computer. A computer-readable storage medium comprising instructions that cause the computer on which they are executed to run a Ver drive to at least one of the Claims 1 - 13 to execute.

Images