EP4294696A1 - Hindernisdetektion im gleisbereich auf basis von tiefendaten - Google Patents
Hindernisdetektion im gleisbereich auf basis von tiefendatenInfo
- Publication number
- EP4294696A1 EP4294696A1 EP22721037.4A EP22721037A EP4294696A1 EP 4294696 A1 EP4294696 A1 EP 4294696A1 EP 22721037 A EP22721037 A EP 22721037A EP 4294696 A1 EP4294696 A1 EP 4294696A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- image data
- rail vehicle
- depth
- rails
- pixel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B61—RAILWAYS
- B61L—GUIDING RAILWAY TRAFFIC; ENSURING THE SAFETY OF RAILWAY TRAFFIC
- B61L23/00—Control, warning or like safety means along the route or between vehicles or trains
- B61L23/04—Control, warning or like safety means along the route or between vehicles or trains for monitoring the mechanical state of the route
- B61L23/041—Obstacle detection
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B61—RAILWAYS
- B61L—GUIDING RAILWAY TRAFFIC; ENSURING THE SAFETY OF RAILWAY TRAFFIC
- B61L15/00—Indicators provided on the vehicle or train for signalling purposes
- B61L15/0072—On-board train data handling
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
Definitions
- the invention relates to a method for detecting obstacles for a rail vehicle.
- the invention also relates to an obstacle detection device.
- the invention also relates to a rail vehicle.
- a supervised object detection with the return of so-called bounding boxes based on machine learning has proven to be quite effective for most marked objects.
- it is quite expensive to create a new one To identify or train a class in a development cycle for a model.
- a database must be sought that includes the relevant obstacles. The data must then either be annotated by the company's own specialist staff or this process must be left to external contractors.
- the new model must be trained and validated accordingly. Then the model has to be implemented in the field and tested in practice.
- the imaged environment of the rail vehicle preferably includes a driving channel of the rail vehicle running in front of the rail vehicle, but can preferably also include peripheral areas to the right and left of the driving channel, for example to identify potential collision obstacles early on before they even get close to the driving channel have arrived.
- 2D image data are generated on the basis of the 3D image data.
- the 3D image data are projected onto a 2D plane, the orientation of which preferably corresponds to the visual perspective of the rail vehicle.
- This visual perspective also preferably includes the perspective of a driver looking forward from the rail vehicle or in the direction of travel and/or the perspective with which at least some of the sensors, which are preferably aligned in the direction of travel or in the direction of the longitudinal axis of the rail vehicle, the Surrounding area detected.
- Farther determines whether a pixel or point is part of an object that protrudes over the terrain plane. This determination is made depending on whether the difference between the depth value of the relevant pixel and the depth value of the rails of the same image segment exceeds a predetermined threshold value.
- a depth value of a pixel of a line-like image segment outside the rails is compared with the respective depth value of the rails inside the line-like image segment.
- the line-like image segment can, for example, comprise a horizontal line in an image.
- all pixels in the tile should have approximately the same depth. In this case, the pixels are all considered part of the terrain layer. Otherwise, it is determined that the pixel is part of an object, possibly an obstacle, which protrudes above the ground level, i.e. in the event that the difference in the depth value of the pixel to the depth value of the rails exceeds a predetermined threshold value.
- the pixel was detected as part of an object that projects beyond the terrain level, depending on a position and/or a detected movement, in particular the speed and direction of movement of the object, it is determined whether the detected object is a represents a potential collision obstacle.
- Annotations for classifications of objects are not strictly required.
- the actual obstacle detection can be carried out, for example, depending on the position of the detected object and by tracking a movement of the detected object. If the detected object moves or even moves in the direction of the track area, the detected object can be classified as a potential obstacle.
- the obstacle detection device has a sensor data interface to an image recording unit for recording three-dimensional image data of a surrounding area of a rail vehicle.
- the three-dimensional image data can be recorded directly by a stereo camera. However, the three-dimensional image data can also initially be recorded by a mono camera, for example from different positions, and a 3D image is then reconstructed on the basis of its 2D image data.
- the obstacle detection device also includes a projection unit for generating 2D image data on the basis of the 3D image data. Part of the obstacle detection device according to the invention is also a localization unit for detecting and localizing rails in the 2D image data.
- the obstacle detection device also includes a depth data determination unit for determining depth data in the 2D image data based on the 3D image data and an assignment unit for dividing the 2D image data into linear image segments, each with a rail section with a constant depth.
- a depth data determination unit for determining depth data in the 2D image data based on the 3D image data
- an assignment unit for dividing the 2D image data into linear image segments, each with a rail section with a constant depth.
- Part of the inventions to the invention obstacle detection device is also a comparison unit for comparing the depth value of a Pixels of a line-like image segment outside the rails with the respective depth value of the rails.
- the obstacle detection device also includes a detection unit for detecting whether the pixel is part of an object projecting beyond the terrain level, depending on whether the difference between the depth value of the pixel and the depth value of the rails of the same image segment exceeds a predetermined threshold value.
- the obstacle detection device also includes an obstacle determination unit for determining, in the event that the pixel was detected as part of an object that projects beyond the terrain level, whether the detected object represents a potential collision obstacle, depending on a position and/or a detected movement of the object.
- the obstacle detection device shares the advantages of the process according to the invention for obstacle detection for a rail vehicle.
- the rail vehicle according to the invention has a sensor unit for capturing 3D image data from the surroundings of the rail vehicle.
- the rail vehicle according to the invention includes the obstacle detection device according to the invention.
- the rail vehicle according to the invention has a control device for controlling a driving behavior of the rail vehicle depending on whether an obstacle in the vicinity of the rail vehicle was detected by the obstacle detection device.
- the rail vehicle according to the invention shares the advantages of the obstacle detection device according to the invention.
- Some components of the obstacle detection device can be designed for the most part in the form of software components. This applies in particular to the sensor data interface, the projection unit, the localization unit, the depth data determination unit, the assignment unit, the comparison unit and the detection unit. In principle, however, these components can also be partially implemented in the form of software-supported hardware, for example FPGAs or the like, particularly when particularly fast calculations are involved.
- the required interfaces for example when it is only a matter of taking over data from other software components, can be designed as software interfaces. However, they can also be in the form of hardware interfaces that are controlled by suitable software.
- a largely software-based implementation has the advantage that even previously existing computer systems in a rail vehicle can be easily retrofitted with a software update after a possible addition of additional hardware elements, such as a video camera, in order to work in the manner according to the invention work.
- the object is also achieved by a corresponding computer program product with a computer program that can be loaded directly into a memory device of such a computer system, with program sections in order to carry out the steps of the method according to the invention that can be implemented using software when the computer program is executed in the computer system will.
- such a computer program product may also include additional components such as documentation and/or additional components, including hardware components such as hardware keys (dongles, etc.) for using the software.
- additional components such as documentation and/or additional components, including hardware components such as hardware keys (dongles, etc.) for using the software.
- a computer-readable medium for example a memory stick, a hard disk or another transportable or permanently installed data medium, on which the program sections of the computer program that can be read and executed by a computer unit, can be used for transport to the storage device of the computer system and/or for storage on the computer system are saved.
- the computing unit can have one or more working micro processors or the like for this purpose.
- the localization of the rails is based on semantic segmentation on deep learning (in German: multi-layer learning, deep learning: relates to a method of machine learning that uses artificial neural networks with numerous intermediate layers between the input layer and output layer uses) carried out based ren.
- a variant of obstacle detection, which can be implemented with the help of machine learning, is based on semantic segmentation.
- semantic segmentation pixels are divided into classes. In a training process, each pixel is annotated. Pixels from railroad tracks are assigned to one class, all other pixels are assigned to a class relating to the background of the image.
- Such a scenario can also be understood as a binary foreground/background scenario.
- the training process can advantageously be limited to two classes, which greatly simplifies the training process.
- the 3D image data are captured by a stereo camera.
- 3D information in particular depth data
- 3D information can be recorded directly from the area surrounding a rail vehicle. While the detection of rails in 2D image data is simplified, 3D information, in particular depth data, is used to detect objects that have a height that exceeds the height of the rails.
- the 3D image data can also be captured dynamically by a mono camera.
- image data is captured from different directions and a 3D image is generated from this image data.
- Such a technology is called “Monodepth” and is described in Clement Godard et al.: “Digging Into Self-Supervised Monocular Depth Estimation” (available on the Internet at the address https://ar-xiv.org/pdf/1806.01260. pdf) described.
- the 3D image data preferably include RGB data as image data that have color differences. Color differences can be used advantageously to differentiate between objects and image sections.
- the depth data is preferably determined by a model based on machine learning.
- the training of such a model can be supervised or unsupervised.
- the obstacles can be detected particularly effectively by removing the terrain plane.
- the depth data of rail points is compared with the depth data of pixels of an image segment in which the rail points are located. If a depth difference exceeds a threshold, the pixels can be classified as object pixels, otherwise they are classified as part of the terrain plane and removed from the image, leaving only object pixels that are processed further.
- a sub-area of the 2D image data can also be defined as a safety area and only the safety area is examined for obstacles. The amount of image data to be evaluated is advantageously reduced, as a result of which the evaluation process is accelerated.
- FIG. 1 shows a flowchart which illustrates a method for detecting an obstacle for a rail vehicle according to an exemplary embodiment of the invention.
- FIG. 2 shows a schematic representation of two-dimensional image data which are generated in the method illustrated in FIG. 1,
- FIG. 3 shows a schematic representation of the recorded image data, with depth data being assigned to the rails pixel by pixel,
- FIG. 4 shows a schematic representation of the image data already shown in FIG. 3, with different objects being detected
- FIG. 5 shows the representation shown in FIG. 3 and FIG. 4 as a representation with the terrain level removed
- FIG. 6 shows the image representation already shown in FIG. 3, FIG. 4, and FIG. 5 again as a representation with the safety area drawn in,
- FIG. 7 shows a schematic representation of an obstacle detection device according to an embodiment of the inventions
- 8 shows a schematic representation of a rail vehicle according to an exemplary embodiment of the invention.
- FIG. 1 shows a flow chart 100 which illustrates a method for detecting obstacles for a rail vehicle according to an exemplary embodiment of the invention.
- step 1.1 three-dimensional colored 3D image data 3D-BD, also referred to as RGB 3D image data, are recorded from the surroundings or a surrounding area U in front of a rail vehicle 81 (see FIG. 8).
- the 3D image data 3D-BD is recorded by a sensor unit 82 arranged on the rail vehicle 81 (see FIG. 8).
- the sensor unit 82 includes an optical camera for recording three-dimensional color image data 3D-BD.
- the 3D-BD image data include information about the topography of the surrounding area U of the rail vehicle 81.
- the surrounding area U comprises a front area in front of the rail vehicle 81, into which the rail vehicle 81 would like to drive.
- step l.II 2D image data 2D-BD are generated on the basis of the 3D image data 3D-BD.
- the 3D image data is projected onto a 2D plane, which is perpendicular to the orientation of the sensor unit 82 .
- step l.III the rails S running in front of the rail vehicle are detected and localized in the 2D image data. In particular, positions P(S) of the rails S are determined.
- step 1.IV depth data T in the 2D image data 2D-BD is also determined on the basis of the 3D image data 3D-BD. This means that depth information assigned to the individual pixels is obtained from the 3D image data in order to determine the depth information of individual pixels in the 2D image data.
- step IV the 2D image data 2D-BD are divided into line-like image segments BS, each with a rail section divided with a constant depth.
- a straight line runs transversely to the rail track, at least in the event that the rail track runs straight. It can then be assumed that, in the case of flat terrain, all points of the transverse straight line have the same depth if there is no object projecting above the terrain plane.
- step 1.VI the depth value T(P) of a pixel P of a linear image segment BS outside the rails S is compared with the respective depth value T(S) of the rails S of this linear image segment BS. If the difference between the depth values T(P), T(S) exceeds a predetermined threshold value SW, then it is concluded that the pixel P belongs to an object 0 rising out of the plane. In this way, pixels are assigned to different objects 0 and the objects are localized.
- the pixel P is not classified as part of an object 0 in step l.VII.
- step l.VIII a check is made as to whether the detected and localized object 0 is within a safety area around the rails S or, if appropriate, is moving in the direction of the rails. If this is the case, the object 0 is classified as a potential collision obstacle KH in step l.VIII. Otherwise, the object 0 is not classified as a collision obstacle KH in step l.VIII.
- the potential collision obstacles are checked, for example by comparing them with map material, in order to be able to distinguish actual collision obstacles KH from objects 0 in the existing infrastructure. Further methods for identifying collision obstacles KH track a detected object 0. If the object 0 moves, in particular in the direction of the track area, it can be classified as a potential collision obstacle KH, if it does not move and is not there either directly in the track area, the detected object 0 can be regarded as harmless.
- FIG. 3 shows a schematic representation 30 of recorded or generated image data 2D-BD, with the rails S being assigned depth data T(S) pixel by pixel. 3 also shows objects in the image, such as a cyclist CL, a mast PL, and buildings BL. FIG. 3 therefore shows a scenario that can be assigned to step 1.IV in FIG.
- FIG. 4 shows a schematic representation 40 of the image data already illustrated in FIG. 3, with different objects BL, CL, PL, such as a cyclist CL, a mast PL and building BL, being detected.
- the pixels of the objects typically have a shallower depth in any image row than the pixels of the rails S.
- the pixel P(PL) in row R ⁇ has a shallower depth T(P(PL)) than the rails S. It can therefore be assumed that the pixel P(PL) belongs to an object that protrudes from the ground plane.
- FIG. 4 therefore shows a scenario which corresponds to steps l.V and l.VI in FIG.
- FIG. 7 shows a schematic representation of an obstacle detection device 70 according to an exemplary embodiment of the invention.
- the trajectory of the detected object 0 is tracked, for example.
Landscapes
- Engineering & Computer Science (AREA)
- Mechanical Engineering (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
Description
Claims
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE102021206475.2A DE102021206475A1 (de) | 2021-06-23 | 2021-06-23 | Hindernisdetektion im Gleisbereich auf Basis von Tiefendaten |
| PCT/EP2022/059110 WO2022268375A1 (de) | 2021-06-23 | 2022-04-06 | Hindernisdetektion im gleisbereich auf basis von tiefendaten |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| EP4294696A1 true EP4294696A1 (de) | 2023-12-27 |
Family
ID=81580403
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| EP22721037.4A Withdrawn EP4294696A1 (de) | 2021-06-23 | 2022-04-06 | Hindernisdetektion im gleisbereich auf basis von tiefendaten |
Country Status (4)
| Country | Link |
|---|---|
| US (1) | US20240326883A1 (de) |
| EP (1) | EP4294696A1 (de) |
| DE (1) | DE102021206475A1 (de) |
| WO (1) | WO2022268375A1 (de) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102023203319A1 (de) * | 2023-04-12 | 2024-10-17 | Siemens Mobility GmbH | Spurgebundene Hinderniserkennung |
| DE102023210109A1 (de) * | 2023-10-16 | 2025-04-17 | Siemens Mobility GmbH | Hindernisdetektion im Gleisbereich durch Kombination von Streckenverlaufsdaten und Bilddaten |
| US12553735B2 (en) * | 2024-01-22 | 2026-02-17 | Disney Enterprises, Inc. | Selective three-dimensional localization and navigation systems and methods |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102014206473A1 (de) | 2014-04-03 | 2015-10-08 | Bombardier Transportation Gmbh | Automatische Assistenz eines Fahrers eines fahrspurgebundenen Fahrzeugs, insbesondere eines Schienenfahrzeugs |
| DE102014222900A1 (de) | 2014-11-10 | 2016-05-12 | Bombardier Transportation Gmbh | Betrieb eines Schienenfahrzeugs mit einem Bilderzeugungssystem |
| JP6307037B2 (ja) * | 2015-03-31 | 2018-04-04 | 公益財団法人鉄道総合技術研究所 | 障害物検知方法およびその装置 |
| JP7062407B2 (ja) * | 2017-11-02 | 2022-05-06 | 株式会社東芝 | 支障物検知装置 |
| JP6843274B2 (ja) | 2018-02-08 | 2021-03-17 | 三菱電機株式会社 | 障害物検出装置および障害物検出方法 |
| DE102020215754A1 (de) | 2020-12-11 | 2022-06-15 | Siemens Mobility GmbH | Optische Schienenwegerkennung |
-
2021
- 2021-06-23 DE DE102021206475.2A patent/DE102021206475A1/de not_active Withdrawn
-
2022
- 2022-04-06 WO PCT/EP2022/059110 patent/WO2022268375A1/de not_active Ceased
- 2022-04-06 EP EP22721037.4A patent/EP4294696A1/de not_active Withdrawn
- 2022-04-06 US US18/574,098 patent/US20240326883A1/en active Pending
Also Published As
| Publication number | Publication date |
|---|---|
| US20240326883A1 (en) | 2024-10-03 |
| WO2022268375A1 (de) | 2022-12-29 |
| DE102021206475A1 (de) | 2022-12-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| DE102016114535B4 (de) | Wegbestimmung für automatisierte Fahrzeuge | |
| DE69322306T2 (de) | Gegenstandserkennungssystem mittels Bildverarbeitung | |
| DE102016107705A1 (de) | Reaktive Pfadplanung für autonomes Fahren | |
| EP4294696A1 (de) | Hindernisdetektion im gleisbereich auf basis von tiefendaten | |
| DE102019123483B4 (de) | Verfahren sowie Kraftfahrzeug-Steuereinheit zum Erfassen einer Umgebung eines Kraftfahrzeugs durch Fusionieren von Sensordaten auf Punktwolkenebene | |
| DE102014206473A1 (de) | Automatische Assistenz eines Fahrers eines fahrspurgebundenen Fahrzeugs, insbesondere eines Schienenfahrzeugs | |
| DE102018215055A1 (de) | Verfahren zum Bestimmen einer Spurwechselangabe eines Fahrzeugs, ein computerlesbares Speichermedium und ein Fahrzeug | |
| DE102018133457B4 (de) | Verfahren und System zum Bereitstellen von Umgebungsdaten | |
| EP4198894B1 (de) | Kamerabasierte fahrzeugkonturerfassung für fahrzeugbehandlungsanlagen | |
| DE102020118589A1 (de) | Verbesserter fahrzeugbetrieb | |
| DE102020201939A1 (de) | Verfahren und Vorrichtung zur Bewertung eines Bildklassifikators | |
| DE102017123226A1 (de) | Verfahren zum Bestimmen einer kritischen Höhe eines vorausliegenden Streckenabschnitts für ein Fahrzeug, das ein Zugfahrzeug und einen Anhänger umfasst | |
| AT506964A4 (de) | Verfahren und vorrichtung zur bilderkennung von bewegten objekten, wie etwa in förderanlagen | |
| DE102023123012B3 (de) | Verfahren und Vorrichtung zur schnellen Kollisionserkennung bei hochautomatisiertem Testen von Fahrvorgängen | |
| DE102018100667A1 (de) | Computersichtvorfusion und räumlich-zeitliche Verfolgung | |
| DE102020005763A1 (de) | Verfahren zur Steuerung eines Fahrzeugs | |
| DE102022208187A1 (de) | Hinderniserkennung im Gleisbereich durch Anomaliedetektion | |
| WO2022042903A1 (de) | Verfahren zur erkennung dreidimensionaler objekte, computerprogramm, maschinenlesbares speichermedium, steuergerät, fahrzeug und videoüberwachungssystem | |
| DE102018109680A1 (de) | Verfahren zum Unterscheiden von Fahrbahnmarkierungen und Bordsteinen durch parallele zweidimensionale und dreidimensionale Auswertung; Steuereinrichtung; Fahrassistenzsystem; sowie Computerprogrammprodukt | |
| WO2023247356A1 (de) | Verfahren und assistenzsystem zum unterstützen einer fahrzeugführung basierend auf einem fahrschlauch und einer begrenzungsschätzung und kraftfahrzeug | |
| EP4381465A1 (de) | Verfahren und kartenerzeugungseinrichtung für ein fahrzeug zur erzeugung einer hochauflösenden karte einer bodenfläche in einem fahrzeugumfeld | |
| DE102004047122A1 (de) | Verfahren zum Überwachen eines Fahrzeugs | |
| EP4232341B1 (de) | Selbstlernendes verfahren zur hinderniserkennung für ein schienenfahrzeug, hinderniserkennungseinrichtung, und schienenfahrzeug mit einer hinderniserkennungseinrichtung | |
| EP4113392B1 (de) | Verfahren zum prüfen der zuverlässigkeit einer ki-basierten objekt-detektion | |
| WO2025082652A1 (de) | Hindernisdetektion im gleisbereich durch kombination von streckenverlaufsdaten und bilddaten |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: UNKNOWN |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |
|
| PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
|
| 17P | Request for examination filed |
Effective date: 20230921 |
|
| AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR |
|
| DAV | Request for validation of the european patent (deleted) | ||
| DAX | Request for extension of the european patent (deleted) | ||
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE APPLICATION IS DEEMED TO BE WITHDRAWN |
|
| 18D | Application deemed to be withdrawn |
Effective date: 20251101 |