Obstacle detection in rail transport

Abstract

The invention relates to a method for a rail vehicle (T) to improve decision-making in the context of obstacle detection. An internal rail vehicle system (SYS TRAIN) performs object detection using first information relating to the environment of the rail vehicle (T), and based on this object detection, a first hazard assessment is carried out regarding the safety relevance of a detected object (K). A query is then sent to an external rail vehicle system for a second hazard assessment using further information relating to the environment of the rail vehicle (T), whereupon a decision is made regarding the course of action to be taken in response to the object (K), taking into account the first and / or the second hazard assessment.Furthermore, the invention relates to a corresponding device or system for data processing, a rail vehicle (T), a computer program, a computer-readable data carrier, and a transmission signal.

Description

[0001] The invention relates to a method for obstacle detection for a rail vehicle.

[0002] In rail vehicles, obstacle detection during movement plays a major role in the safe operation of the vehicle. Obstacles can be, for example, fallen trees or living creatures on the tracks. Various technologies are used to detect the presence of obstacles on or near the tracks and, in response to a detected obstacle, take appropriate action to prevent dangerous situations or collisions. For example, there is track monitoring, for which sensors are installed on the rails to monitor the condition of the track. Furthermore, trains can be equipped with signal detection systems that detect and interpret signals on the track. By monitoring the signals in this way, a train can determine whether it has a clear path or whether a signal is red, which could indicate an obstacle.

[0003] Due to the high safety relevance of reliable obstacle detection, we are continuing to seek improvements in this area. In particular, we are trying to eliminate the possibility of obstacles being overlooked due to human error, particularly due to driver inattention.

[0004] The invention is based on the object of demonstrating an improved method for obstacle detection for a rail vehicle.

[0005] This object is achieved by a method having the features of claim 1. Furthermore, the invention relates to a corresponding device or system for data processing, a rail vehicle, a corresponding computer program, a corresponding computer-readable, preferably non-volatile, storage medium, and a corresponding transmission signal. Advantageous embodiments and further developments are the subject of subclaims.

[0006] The method according to the invention serves to improve decision-making within the framework of obstacle detection for a rail vehicle. A system within the rail vehicle first performs object detection using first information relating to the surroundings of the rail vehicle. Then, based on the object detection, a first risk assessment is carried out with regard to the safety relevance of a detected object. Subsequently, the system within the rail vehicle sends a request to a system outside the rail vehicle for a second risk assessment using second information relating to the surroundings of the rail vehicle. Finally, the system within the rail vehicle makes a decision regarding a course of action to be taken in response to the object, taking into account the first and / or second risk assessment.

[0007] The initial hazard assessment is performed within the rail vehicle based on initial information about the rail vehicle's surroundings. This hazard assessment relates to a detected object located in the rail vehicle's surroundings; this object can be located on or near the tracks. The hazard assessment indicates whether and, if so, to what extent the object poses a hazard, allowing a decision to be made regarding an appropriate response to the hazard.

[0008] The validity and usability of the first risk assessment depends in particular on the quality of the initial information. If this information describes the rail vehicle's surroundings well, it may be sufficient to base a decision on how to respond to a detected object on the first risk assessment alone. However, to obtain a better risk assessment, it is advisable to generate it using additional information about the rail vehicle's surroundings. For this purpose, the system on the rail vehicle sends a request to a system external to the rail vehicle. The external system thus represents an escalation level that can provide additions and / or corrections to the results of the system on the rail vehicle. In accordance with the request, the external system provides a second risk assessment.The on-board system can then base its decision on how to respond to the object on the first hazard assessment, the second hazard assessment, or both. Both the question of whether to involve the off-board system and the question of which hazard assessment to base the decision on is a decision made by the on-board system.

[0009] Preferably, the first information relating to the surroundings of the rail vehicle comprises measured values ​​acquired by sensors on the rail vehicle. Cameras, lidar, and radar sensors are particularly suitable for this purpose. These can be specifically intended for use in the invention or alternatively used for other purposes.

[0010] Object detection can be performed using an AI system. Such a system is trained with training data corresponding to the initial information to detect and, if necessary, classify objects.

[0011] The result of object recognition can be: a classification of the object, and / or a size of the object, and / or a position of the object, and / or an indication of the reliability of object recognition, i.e. confidence.

[0012] Preferably, all of this information is used as the basis for the initial risk assessment. Depending on the object detection algorithm used, additional information about a detected object may be output.

[0013] The hazard assessments – both the first and second hazard assessments – indicate the degree of hazard to the rail vehicle and / or to persons inside the rail vehicle and / or to the object and / or to another object, preferably located in the immediate vicinity of the rail vehicle. This can be expressed as a binary value, i.e., whether a hazard is present or not. Alternatively, a scale indicating the level of hazard is possible, e.g., from 1 to 10.

[0014] It is advantageous if, before the request is sent to the system external to the rail vehicle, a decision is made about sending the request based on the first risk assessment and current information about the rail vehicle, in particular the position and / or route of the rail vehicle. Once the first risk assessment is available, a decision-making process is carried out within the rail vehicle which determines whether the system external to the rail vehicle should be included. Taking current information about the rail vehicle into account makes it possible, for example, to identify a particular urgency due to a particular position of the vehicle, which would argue against including the system external to the rail vehicle. Furthermore, the current information about the rail vehicle can also be used to estimate the likelihood of success of a risk assessment by the external system, e.g.B. if it is known that the external system can use certain sensors whose coverage area includes the position of the detected object.

[0015] The second risk assessment by the system external to the rail vehicle can be performed taking into account the object detection of the system internal to the rail vehicle. In this way, the object detection already performed by the system internal to the rail vehicle can be combined with the second information to enable improved object detection in the system external to the rail vehicle. This then results in a correspondingly improved risk assessment. In addition to the object detection of the system internal to the rail vehicle, the second risk assessment can also consider further information from the system internal to the rail vehicle.

[0016] The decision regarding the course of action can be made based on only the second risk assessment, provided it is provided within a specified time. For this purpose, the on-board system can, for example, use a timer. After the request is sent to the off-board system, the first risk assessment is used if the second risk assessment was not provided by the off-board system. Conversely, if the second risk assessment is received before the timer expires, this second risk assessment can be used for the decision.

[0017] The procedure to be implemented in response to the object preferably includes braking the rail vehicle. This can prevent a collision with the object. Other measures, such as darkening the windows or activating a microphone, can be used alternatively or in addition.

[0018] The method according to the invention and / or one or more functions, features and / or steps of the method according to the invention and / or one of its embodiments can be carried out with computer support. It can be carried out or implemented, for example, using one or more computers, processors, application-specific integrated circuits (ASICs), digital signal processors (DSPs) and / or so-called "field programmable gate arrays" (FPGAs). It can also be carried out at least partially in a cloud and / or in an edge computing environment. One or more interacting computer programs are used for the computer-supported process. If multiple programs are used, they can be stored and executed together on one computer, or on different computers at different locations.Since this is functionally equivalent, the terms “the computer program” and “the computer” are formulated in the singular.

[0019] The invention is explained in more detail below using an exemplary embodiment. The following shows: Fig. 1: a rail vehicle with an obstacle detection system, Fig. 2: Processes and components of the obstacle detection system, as well as processes and components of an external system, Fig. 3: a flow chart, Fig. 4: an obstacle detection system for a rail vehicle.

[0020] As the level of automation in rail transport continues to increase, solutions are being sought that will automate some of the train driver's tasks, or even eliminate them altogether. The focus here is particularly on the detection of obstacles on the track. The size and type of the obstacle, as well as its location in relation to the track being traveled, are crucial for the risk assessment. Living beings, in particular, are endangered by the train on the one hand, and can also pose a danger to the train on the other. In response to the detection of a hazardous situation, braking is usually initiated and / or a warning signal is issued via a microphone.

[0021] Fig. Figure 1 shows a rail vehicle T with a SYS TRAIN system for obstacle detection. Currently, obstacles are usually detected by the train driver looking out of the window F of the train. An obstacle is defined as an object located on or near the tracks G that poses a threat to the safe operation of the rail vehicle T or that could be endangered by the movement of the rail vehicle T. An example is shown in Fig. 1 shows a child K running near the tracks G, which represents such an obstacle.

[0022] Obstacle detection is performed automatically by the SYS TRAIN system, as described below. For this purpose, the rail vehicle T is equipped with SENS sensors, which can be, for example, cameras, radar sensors, or lidar sensors. Unlike in Fig. 1 - should not be located at the front center of the traction vehicle. The processes and components of the SYS TRAIN system for obstacle detection are described in Fig. 2. The SYS TRAIN system is not only responsible for detecting obstacles, but also for classifying and evaluating them, for deciding how to deal with a detected hazardous situation, and for taking measures to avert damage.

[0023] The initial input data is the measured values ​​SENS MEAS from the SENS sensors. These are made available to an AI system KI, which detects and classifies obstacles. Known methods can be used for this, such as those developed for the automotive sector. Based on the input SENS MEAS from the sensors, the AI-based obstacle detection system analyzes the surroundings of the rail vehicle T and creates an object model that lists the objects detected in the track area and the immediate vicinity. This object model contains the position of the objects relative to the train and track area, their size, and the respective classification of the objects by category, e.g., person, animal, object. In addition, information on the confidence or reliability of the respective classification, as well as other parameters of the object such as size and position, can be provided.

[0024] The ASSESS component performs a risk assessment based on the created object model. This serves to determine whether a safety-relevant situation exists. A situation is particularly safety-relevant if it involves a danger to a person or other living being, both in the track area and in the rail vehicle. Also safety-relevant are situations that could lead to damage and / or derailment of the rail vehicle, the latter of which can also involve danger or injury to passengers and personnel on the rail vehicle.

[0025] System context information (KON) can be included in the risk assessment. This is provided by the train control system (ZS). This is the control system of the rail vehicle, which serves both to provide the system context information (KON) and, via an interface to the DECIDE component (explained later), also implements actuator-based decisions of the action strategy, such as activating a warning tone and / or initiating braking. The system context information (KON) includes data such as the date, time of day, route information, as well as rail vehicle-specific parameters such as the weight and the current speed and position of the rail vehicle.The route information, combined with the current position of the rail vehicle, can be used to determine whether the rail vehicle is near a tunnel, a level crossing, a station, or similar hazardous areas. Rail-vehicle-specific parameters such as vehicle weight and current speed can be used to determine the braking distance required to bring the vehicle to a standstill. Additional parameters such as track gradient, braking force build-up time, and braking performance can be taken into account.

[0026] The goal of the risk assessment is therefore to make a decision about whether or not a safety-relevant situation exists. However, this can lead to false-positive assessments (i.e., the assessment of a non-hazardous situation as hazardous), as well as false-negative assessments (i.e., the assessment of a hazardous situation as not hazardous). On the one hand, the AI-based obstacle detection may already have detected the object with only low confidence. On the other hand, the assessment of the train's surroundings may not be appropriate due to the limited information.Making a decision about the presence or absence of a safety-relevant situation based solely on the sensors on the train and the AI-based automated evaluation system is not sensible if a reliable decision cannot be made in a specific situation based solely on the information available in the rail vehicle. In this case, both false-positive and false-negative decisions are undesirable. In the former case, the decision that a safety-relevant situation exists even though this is not actually the case, and in the latter case, the decision that no safety-relevant situation exists even though this is actually the case.

[0027] To ensure sufficient safety, it would be possible to implement obstacle detection, for example, using multiple artificial intelligence systems. If there is a discrepancy, a hazardous situation could be assumed. While this would result in a high level of safety, it could severely limit the availability of the rail vehicle due to insufficient precision or an excessively high fault-positive rate, for example, by initiating braking operations too frequently.

[0028] Therefore, the aim is to incorporate information from other sensors not located on the rail vehicle into the decision-making process, e.g., from cameras on track systems or stations, or from sensors on other rail vehicles. The goal is to use this additional external information to increase the reliability of decision-making regarding the existence of a hazardous situation.

[0029] The STRATEGY component decides on an escalation strategy. This is intended to take into account the aforementioned possibilities for incorporating external information in order to improve decision quality and thus operational availability. The escalation strategy specifies whether and how external information should be used. The decision on the escalation strategy to be pursued is made based on the risk assessment performed in the ASSESS component, which also includes the existing estimate of the confidence of the object classification results. Furthermore, the system context information (KON) can also be included.

[0030] Escalation strategies include, for example: • Delay in the decision on the risk assessment while external information is being requested. • Immediate decision on the risk assessment, but still request external information to support the decision, and after receiving the external information, re-examine and, if necessary, revise the decision. In this regard, this escalation strategy should specifically specify the weight to be given to the two risk assessments, especially if a decision from a land-based system conflicts with local information from the rolling stock. • Setting a timer within which external information must be available; otherwise, the risk assessment is decided based solely on the local information of the rail vehicle. • Involve train crew to obtain specific information about the situation or instructions. • No query of external information, but decision on the risk assessment based solely on the local information of the rail vehicle.

[0031] The main decision regarding the escalation strategy is whether or not to involve an external system. If involved, questions regarding timing and prioritization follow.

[0032] Fig. Figure 2 shows the internal processes and components of the SYS TRAIN system within the rail vehicle, as well as the processes and components of an external SYS REMOTE system. The external SYS REMOTE system can be located, for example, in a signal box, control center, or operations center. A decentralized location of the external SYS REMOTE system is also possible, for example, in a cloud-based environment.

[0033] The HANDLING component receives requests from the STRATEGY component of the rail vehicle and orchestrates the further processing sequence in the external system. Communication between the internal rail vehicle system SYS TRAIN and the external system SYS REMOTE takes place via radio. With the request from the STRATEGY component, the risk assessment performed in the ASSESS component is forwarded and an external risk assessment is requested. The HANDLING component analyzes this request and checks which external information could support the risk assessment. A corresponding query for external information is forwarded to the information supplementation module INFO MODULE. The information supplementation module INFO MODULE determines whether the relevant information for the risk assessment is available. This information is obtained from external sensors EXT SENS 1, EXT SENS 2, ..., EXT SENS n, e.g.Cameras on platforms, tunnels, and bridges, or sensors on the tracks, e.g., for wayside train monitoring, are extracted. Information about known obstacles on the track, such as fallen trees, or weather data can be used as external information in addition to or instead of measurements. The latter can indicate that obstacles on the tracks are to be expected, for example, due to severe weather. Finally, information from other rail vehicles can also be used, e.g., if they have detected and reported obstacles or damage to the vehicle or tracks. The external information is made available to the EXT ASSESS component for external evaluation of the risk assessment.

[0034] For the external risk assessment, the risk assessment performed internally in the ASSESS component is combined with the external information. In the EXT ASSESS component, a comprehensive risk assessment can be performed based on external and internal information, which is then made available to the requesting system, SYS TRAIN. In the requesting system, SYS TRAIN, this external risk assessment is received by the DECIDE component. This component is a decision-maker that decides on an action strategy and implements it. Implementation occurs via an instruction to the train control system (ZS), which then controls the corresponding actuators or triggers processes.

[0035] Fundamental to the decision regarding the actions to be taken is the escalation strategy issued by the STRATEGY component, which is binding for the DECIDE component. If, according to this escalation strategy, for example, after a timer has expired, no further waiting for input from the external system is necessary, an action strategy must be developed based on the internal risk assessment of the ASSESS component. However, if, for example, according to the escalation strategy, the external risk assessment is fundamentally decisive, an action strategy must be developed based on the external risk assessment of the EXT ASSESS component. If, for example,If, according to the escalation strategy, a query to the external system is not advisable due to time constraints because the obstacle identified based on the internal information is located a short distance in front of the rail vehicle, an action strategy must be developed based on the internal risk assessment of the ASSESS component. It is also possible that, together with the external risk assessment, the external system SYS REMOTE issues an instruction that this assessment must be adopted; in this case, the escalation strategy would be corrected if necessary by this instruction.

[0036] The decision-maker DECIDE thus decides on the course of action based on the selected escalation strategy, based on the information received from the external system and / or the internal risk assessment, whereby the system context information CON may also be taken into account. The latter could, for example, indicate that the rail vehicle is in a tunnel, so braking to a standstill would currently be unfavorable.

[0037] A two-level risk assessment is therefore introduced: this is divided into the level of the internal system of the rail vehicle and the land-based escalation level, which represents the external system. The escalation strategy determines the roles of the two levels. The decision-making authority regarding when and to what extent the land-based system should be involved lies with the internal system of the rail vehicle. With a single-level approach, however, the information available at the other level could not be used effectively, which is disadvantageous. Using both levels, however, offers several advantages: • Greater availability of the rail vehicle in unclear hazard situations through escalation to the external system due to an improved hazard assessment of the situation by the external system. • By selecting one of the available escalation strategies, it is ensured that the most appropriate approach is selected and pursued under various relevant aspects such as the expected response time of the external system or the prospects of success of the query of the external system in the current context. • The option to escalate to the external system allows an internal automated function for object detection and risk assessment to be configured to a sufficiently high security level, knowingly accepting an increased proportion of false-positive alarms. These false-positive cases can be corrected by the external system. • Time monitoring using a timer ensures that a timely and reliable response can be made even if the external system is unavailable.

[0038] In summary, Fig. 3 a flowchart of the described procedure, with only the most important steps shown. First, an internal risk assessment GFB INT is prepared in the internal system SYS TRAIN based on the local information of the rail vehicle. Based on this and on system context information that describes the current state of the rail vehicle, in particular with regard to its position and braking capability, a decision is made on the escalation strategy ESK to be used. If this involves the involvement of the external system SYS REMOTE, the internal system SYS TRAIN requests the preparation of a risk assessment GFB EXT of the external system SYS REMOTE. Taking this external risk assessment GFB EXT into account, the internal system SYS TRAIN can then decide on the action ACTION to be carried out and initiate this action.If the escalation strategy ESK does not include the inclusion of the external system SYS REMOTE, the decision on the action to be taken ACTION is made based solely on the internal risk assessment GFB INT.

[0039] Fig. Figure 4 shows how the internal SYS TRAIN system can be structured. While the components explained in more detail below are shown as a single component in the figure, it is also possible for them to be present in multiple versions, e.g., as a distributed system. In this way, the functionality of the SYS TRAIN system can be distributed among several, possibly hierarchically linked, systems. This system or systems can be located in the rail vehicle.

[0040] The SYS TRAIN system comprises a computing unit or processor PRO. This is connected to a memory MEM in which a computer program PROGRAM is stored. The memory MEM is preferably a non-volatile computer-readable data storage medium. Storage can be implemented in any way that is suitable for ensuring readability by a computing unit, such as magnetic storage, e.g., using a floppy disk, optical storage, e.g., using a CD, magneto-optical storage, ROM (Read Only Memory), RAM (Random Access Memory), EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), or flash memory.

[0041] By executing the instructions of the PROGRAM program in the PRO processor, the steps of the procedure described above are carried out. In particular, objects are detected and classified, a risk assessment is performed, a decision is made regarding the escalation strategy, and a decision is made regarding the action strategy based on the risk assessment. For this purpose, the PRO processor is connected to an IN / OUT input and output unit, via which information can be exchanged between the SYS TRAIN system and other components and / or a user. This interface can be designed in a suitable manner, preferably via radio, and communication can take place using suitable standards. A request for an external risk assessment can be sent and a requested external risk assessment can be received via the IN / OUT input and output unit.

[0042] The invention has been described above using an exemplary embodiment. It is understood that numerous changes and modifications are possible without departing from the scope of the invention.

Claims

[1] Method for a rail vehicle (T) to improve decision-making in the context of obstacle detection, in which an internal rail vehicle system (SYS TRAIN) object detection is carried out using initial information (SENS MEAS) concerning the surroundings of the rail vehicle (T), based on the object detection, an initial risk assessment is carried out with regard to the safety relevance of a detected object (K), a request is sent to a system external to the rail vehicle (SYS REMOTE) after a second risk assessment using second information concerning the environment of the rail vehicle (T), a decision is made on a course of action to be taken in response to the object (K), taking into account the first and / or second risk assessment. [2] Method according to claim 1, wherein the first information (SENS MEAS) relating to the surroundings of the rail vehicle (T) comprises measured values acquired by sensors (SENS) of the rail vehicle (T). [3] Method according to claim 1 or 2, wherein the object recognition is carried out using an AI system. [4] Method according to one of claims 1 to 3, in which the result of the object recognition is: a classification of the object (K), and / or a size of the object (K), and / or a position of the object (K), and / or an indication of the reliability of the object recognition. [5] Method according to one of claims 1 to 4, wherein the risk assessments indicate a level of risk for the rail vehicle (T) and / or for persons within the rail vehicle (T) and / or for the object (K) and / or for another object. [6] Method according to one of claims 1 to 5, in which, before the request to the system external to the rail vehicle (SYS REMOTE), a decision is made on sending the request based on the first risk assessment and current information concerning the rail vehicle (T), in particular the position and / or the route of the rail vehicle (T). [7] Method according to one of claims 1 to 6, in which the second risk assessment is carried out by the system external to the rail vehicle (SYS REMOTE) taking into account the object detection of the system internal to the rail vehicle (SYS TRAIN). [8] A method according to any one of claims 1 to 7, wherein the decision on the course of action to be taken is made based on only the second risk assessment, provided that it is made available within a predetermined time. [9] Method according to one of claims 1 to 8, wherein the procedure to be carried out in response to the object (K) comprises braking the rail vehicle (T), and / or darkening windows and / or activating a microphone. [10] Device or system for data processing (SYS TRAIN), comprising means for carrying out the method according to one of claims 1 to 9. [11] Rail vehicle (T) with a device according to claim 10, with Sensors (SENS) for detecting the first information (SENS MEAS) concerning the surroundings of the rail vehicle (T), and Actuators for implementing the action to be taken in response to the object (K) according to the decision. [12] Computer program (PROGRAM) comprising instructions which, when executed by a computer, cause the computer to carry out the steps of the method according to any one of claims 1 to 9. [13] Computer-readable storage medium (MEM) with a computer program (PROGRAM) according to claim 12. [14] Transmission signal transmitting the computer program (PROGRAM) according to claim 12.

Images