Device for the reliable detection of the absence of persons in a monitored area

Abstract

Device (100) for the reliable detection of the absence of persons in a monitored area of ​​a technical installation, comprising: a data acquisition unit (102) configured to record sensor data relating to an environment of the technical installation, provided by at least one sensor, a feature extraction unit (104) configured to extract human-specific features from the sensor data, and a deductive decision module (106) configured to initiate a safety-related action if, based on a deductive conclusion derived from the extracted human-specific features, the presence of a person cannot be ruled out.

Description

[0001] The present invention relates to a device for the reliable detection of the absence of persons in a monitored area of ​​a technical plant and to a corresponding method for this purpose.

[0002] Reliable person detection has proven to be a particular challenge in security technology for years. The main difficulty lies in the fact that while existing systems and sensors are capable of detecting moving objects, they cannot reliably distinguish between different types of objects, such as automated guided vehicles (AGVs) and people. Therefore, generalized methods are often used that are not specifically designed for person detection. These methods are based on detecting movement and presence in a monitored area, but without identifying the type of object.

[0003] Consequently, rigid safety concepts are applied to minimize the risk to operators at technical facilities. Such concepts are often conservative and define strict safety zones in which machines automatically stop or slow down as soon as movement is detected. This leads to a reduction in machine productivity and availability, as these safety measures cause frequent interruptions and delays. Furthermore, rigid safety concepts hinder effective collaboration between humans and machines.

[0004] The report by the Federal Institute for Occupational Safety and Health (BAuA) (A. Richter, "Reliable Person Detection in Human-Machine Interaction," BAuA: Report Compact, 1st edition. Dortmund: Federal Institute for Occupational Safety and Health, 2017. Pages: 3, Project number: F 2322) examines this issue in detail and investigates various systems for automatic person detection. It highlights that different technologies, such as 3D cameras, thermal imaging cameras, radar systems, and pressure-sensitive floor coverings, each have specific advantages and disadvantages. None of these systems alone can fully meet the requirements for precise and reliable person detection. For example, 3D cameras and multi-camera systems are well-suited for determining the position of people, while thermal imaging cameras are useful in poorly lit environments, and radar-based systems enable the detection of concealed individuals.

[0005] However, all these technologies have their limitations. Ultrasound systems, for example, are less suitable for precise positioning due to their low angular resolution. Similarly, while 2D laser scanners and pressure-sensitive floor coverings can detect movement, they cannot distinguish between different objects, making them unsuitable for direct human-machine interaction.

[0006] The solution proposed in the report is the use of multi-sensor systems. By combining different sensors, the strengths of each system can be leveraged and their weaknesses compensated for. Such systems offer redundancy and complementarity, leading to more reliable and accurate person detection. Redundancy means that failures or erroneous measurements of individual sensors can be compensated for by other sensors, while complementarity ensures that different sensor types capture various characteristics and aspects of the environment, resulting in a more comprehensive analysis. The challenge, however, lies in integrating these complex systems into practice and ensuring their reliable operation under diverse operating conditions.This integration requires careful coordination and calibration of the various sensors, as well as robust software for processing and interpreting the combined data.

[0007] Even if multiple sensors are successfully integrated, 100% certainty that a person will be detected cannot be achieved. Factors such as unexpected environmental conditions, technical malfunctions, or human error can affect detection accuracy.

[0008] Against this background, the purpose of this disclosure is therefore to specify an alternative approach to reliable person detection in security technology. In particular, it aims to develop alternative approaches for the reliable detection of persons in monitored areas of technical installations that overcome the limitations and uncertainties of existing technologies and improve the reliability and security of the detection.

[0009] This task is solved by a device for the reliable detection of the absence of persons in a monitored area of ​​a technical plant, comprising: a data acquisition unit configured to record sensor data relating to an environment of the technical plant, provided by at least one sensor; a feature extraction unit configured to extract human-specific features from the sensor data; and a deductive decision module configured to initiate a safety-related action if, based on a deductive conclusion derived from the extracted human-specific features, the presence of a person cannot be ruled out.

[0010] The idea is therefore to replace reliable person detection with reliable non-person detection. This approach is based on a reversal of conventional logic in automation technology. While existing technology attempts to identify people with the highest possible degree of certainty, this approach proposes not to rule out the presence of people. This method uses deductive reasoning, applying general rules to specific cases to draw more reliable and accurate conclusions, provided the premises are correct.

[0011] Deductive reasoning uses clearly defined rules and collected sensor data to determine whether the presence of a person cannot be ruled out. This means the system does not attempt to definitively prove that an object is a person, but rather to determine whether an object could possibly be a person – that is, whether a person's presence cannot be excluded. If the sensors detect individual characteristics of a person, it is assumed that a person's presence cannot be excluded, and appropriate safety measures are taken.

[0012] This approach has several advantages. First, it increases security because it is conservative and only initiates protective measures when uncertainty exists. Second, it improves the flexibility and adaptability of security systems, as they are not dependent on the error-prone and often inadequate technology for unambiguous person identification. Third, this approach enables more efficient integration of multi-sensor systems. The various sensors can be combined to create a broader data foundation upon which deductive inferences can be based.

[0013] Specifically, the claimed solution provides a device for the reliable detection of the absence of persons in a monitored area of ​​a technical plant, comprising three essential components: a data acquisition unit, a feature extraction unit, and a deductive decision module.

[0014] The data acquisition unit can be coupled with a sensor and serves to collect sensor data from the environment. This sensor data can take various forms, such as visual data from cameras, thermal images, radar signals, or other sensor types suitable for detecting objects and their properties.

[0015] The feature extraction unit is responsible for extracting human-specific features from the recorded sensor data. These features can include various physiological or behavioral indicators, such as the contours of the human body, movement patterns, body temperature, heartbeat, or other characteristic features that indicate the presence of a human.

[0016] The deductive decision module uses the extracted features to determine, through deductive reasoning, whether the presence of a person in the monitored area cannot be ruled out. If, based on the detected features, there is a possibility that a person is in the monitored area, the module initiates a safety-related action. This action could include, for example, stopping or slowing down a machine, triggering an alarm, or other measures to ensure safety.

[0017] A practical example would be the combination of 3D cameras, thermal imaging cameras, and radar systems, along with a feature extraction unit that extracts human-specific features from the sum of the sensor data. If a feature indicates the presence of a person, their presence is assumed, and safety measures, such as stopping a machine or changing its operating mode, are activated.

[0018] Overall, the approach of secure non-person detection offers a promising alternative to traditional probability-based person detection by taking into account the inherent uncertainties of current technologies and relying on deductive reasoning, which enables more robust and secure decisions.

[0019] In a further embodiment, the deductive decision module can be an inference machine.

[0020] An inference engine uses predefined rules and logical reasoning to make decisions, meaning the results are transparent and traceable. This leads to more robust security measures, as the inference engine systematically analyzes all available data and operates based on clear, logical rules.

[0021] In a further development, the deductive conclusion can involve linking the extracted human-specific characteristics with a logical OR (ORing).

[0022] By using the logical OR operator, various human-specific characteristics are combined in such a way that the presence of just one of these characteristics is sufficient to rule out the presence of a person. This means that if even one of the characteristics, such as movement patterns, body contour, or temperature, is detected, the conclusion is drawn that a person may be in the monitored area. This method simplifies the deductive process because it does not rely on the simultaneous presence of all characteristics, but considers any single characteristic sufficient. This makes the detection logic more robust and less prone to error, thus increasing the security and reliability of the monitoring systems.

[0023] In a further embodiment, the feature extraction unit can be configured to extract and analyze several human-specific features simultaneously.

[0024] By simultaneously capturing and analyzing multiple human-specific characteristics such as contours, movement patterns, temperature, and heart rate, the strengths of different sensors are combined. This enables more robust and reliable detection, as the weaknesses of one sensor can be compensated for by the strengths of another. For example, a camera can provide visual information, while a radar sensor detects movement regardless of lighting conditions, and a thermal imaging camera reveals temperature differences.

[0025] The ability to extract and analyze multiple features simultaneously not only simplifies the detection process but makes it possible in the first place. Without sensor fusion, many scenarios would not be adequately covered, as individual sensor types each have their specific limitations. By fusing sensor data, deductive inferences can be drawn with greater accuracy and reliability, increasing the safety and effectiveness of the entire system. This combination of data leads to a more comprehensive picture of the environment and makes it possible to rule out the presence of a person based on a broader data set.

[0026] In a further embodiment, the feature extraction unit can be a AI-based feature extraction unit that uses a trained neural network for feature extraction.

[0027] A AI-based feature extraction unit is a system designed to identify and extract relevant features from raw data. This unit utilizes a trained neural network to recognize and analyze complex patterns in the data. By employing deep learning algorithms, the system can learn from large amounts of training data and identify specific features such as contours, motion patterns, or temperature differences. The feature extraction unit then operates in real time during inference, delivering precise and reliable information that can be used for further decision-making. The neural network can consist of multiple layers of neurons that work together to identify and extract the relevant features, with each layer processing specific aspects of the data.The first layer can recognize simple features like edges, while deeper layers identify more complex features such as faces or human movements. Through continuous training and adaptation, the neural network can further improve its accuracy and efficiency over time, making it suitable for various applications such as automatic person detection or security area monitoring.

[0028] In a further embodiment, the sensor can include a camera, in particular a 3D camera or a thermal imaging camera.

[0029] Cameras in general, and 3D cameras in particular, can capture spatial information and create accurate models of the environment, while thermal imaging cameras can detect temperature differences that indicate human presence. These technologies can complement each other to provide a robust foundation for feature extraction.

[0030] In a further embodiment, the sensor can include a radar sensor.

[0031] Radar sensors are particularly versatile and powerful because they can detect not only the position and movement of objects, but also specific features that indicate the presence of people.

[0032] Radar sensors, for example, have the capability of radar cross-section (RCS) analysis. RCS measures the radar energy reflected by objects and can help identify different materials and structures. Humans have characteristic RCS values ​​that differ from those of other objects, enabling reliable detection. In addition to RCS, radar sensors can measure energy density, which provides information about the presence and density of objects in the monitored area.

[0033] Furthermore, radar sensors can also detect subtle movements associated with human vital functions. For example, they can detect a person's heartbeat and respiration by capturing minute movements of the chest. This ability to recognize vital signs such as heartbeat and respiration allows for even more precise identification of people, even when they are not actively moving. These additional detection features increase the reliability of absence detection and contribute to the security and effectiveness of the surveillance system.

[0034] In a further embodiment, the feature extraction unit can be configured for the detection of contours, skeleton, heartbeat, respiration, temperature, energy density, and / or movement patterns.

[0035] These features can be captured either by a single sensor or by multiple sensors, which significantly increases the flexibility and accuracy of the system.

[0036] A single sensor can capture several of these features simultaneously. For example, a sophisticated 3D camera can detect both the contours and the skeleton of a person, while a thermal imaging camera can detect both temperature and respiration based on temperature changes in the chest area. These multifunctional sensors reduce the need for numerous different sensors and simplify system design.

[0037] Alternatively, various characteristics can be captured by different sensors to improve the robustness and precision of detection. A radar sensor could monitor an object's movement patterns and energy density, while an optical sensor captures its contours and skeletal structure. Simultaneously, another sensor, such as an infrared camera, could measure body temperature and even detect the smallest movements associated with breathing or heartbeat. This cross-sensor data acquisition enables more comprehensive analysis and increases the reliability of person detection by leveraging the specific strengths of each sensor and compensating for its weaknesses.

[0038] In a further embodiment, at least one sensor can be intrinsically safe.

[0039] In this context, "intrinsically safe" means that the sensor is designed in such a way that it cannot create hazardous conditions for people or the environment, even under fault conditions. This is achieved through specific design principles that ensure that potential faults, such as short circuits or component failures, do not lead to hazardous situations. Intrinsically safe systems and components are often equipped with redundant mechanisms, integrated self-tests, and continuous monitoring functions to ensure that they operate reliably and safely, even if parts of the system should fail.

[0040] In another embodiment, the feature extraction unit can be implemented in a non-safety-oriented manner.

[0041] According to this design, the feature extraction unit does not need to be implemented in a safety-oriented manner, particularly if the sensor is intrinsically safe and the decision module operates deductively. This offers the advantage that the feature extraction unit can be implemented cost-effectively, as simple development processes can be used.

[0042] In a further embodiment, the technical system can be operated in a first mode in which the technical system poses a danger to persons and objects in the vicinity of the technical system, and in a second mode in which the technical system poses no danger or a manageable danger to persons and objects in the vicinity of the technical system, and wherein the safety-related action includes switching from the first mode to the second mode if the presence of a person in the vicinity cannot be ruled out.

[0043] In this configuration, the technical system can be operated in two different modes. The first mode is, for example, normal operation, in which the technical system functions without restrictions. In this mode, the system's full performance and productivity can be utilized, with all functions and processes running without limitations.

[0044] The second mode is a safe operating state in which the system is not completely shut down, but is configured so that there is no danger to an operator, or any danger is manageable. In this mode, the system is set up so that an operator can interact with it safely. This can be achieved, for example, by reducing the speed, activating additional safety precautions, or restricting certain hazardous functions.

[0045] A key advantage of this approach is that even if a person is mistakenly believed to be present, the system, while switching to a slower mode, does not shut down completely, as would be the case with a rigid safety concept. This means the system remains operational and can continue to perform essential tasks while ensuring operator safety. This flexibility allows the system to operate both efficiently and safely by dynamically adapting to the presence of people in the monitored area without causing unnecessary interruptions or productivity losses.

[0046] In particular, the device may have an additional safety device that is activated in the second mode to ensure final protection.

[0047] This additional safety device ensures that even in safe mode, no uncontrolled hazards exist. It provides an extra layer of protection that continuously monitors and can intervene immediately should a hazardous situation arise despite the reduced operating parameters. This ensures maximum operator safety and guarantees that the system can continue to operate efficiently even under the safest conditions.

[0048] In a further embodiment, the device can also include a test unit which is configured to verify the functionality of the feature extraction unit by means of an agent test.

[0049] The test unit can be used to continuously verify the functionality of the feature extraction unit, particularly the algorithms or neural networks used to recognize human-specific features. For example, agent tests can be performed to ensure that feature recognition functions correctly and that the corresponding algorithms execute properly. A dataset covering a wide range of human-specific features can be provided for this purpose. The feature recognition test is conducted in such a way that the algorithms do not operate with the actual sensor data, but rather with data from this specific test dataset. This test data represents various scenarios and features that can occur in the real-world environment.

[0050] Agent tests can be performed at the request of a safety controller. The safety controller initiates the tests and monitors the results to ensure that the feature extraction unit and its algorithms are still functioning correctly. The expectation for these tests is clearly defined: The algorithms must correctly identify and process the features from the test data. This approach ensures that the feature extraction unit is not only functional but also continues to deliver the expected results.

[0051] Regularly performing these agent tests increases the reliability and safety of the overall system. Should a test fail, the safety controller can immediately take measures to minimize potential risks, such as switching to a safe operating mode or triggering maintenance procedures. This continuous testing mechanism ensures that the feature extraction unit always operates optimally and that the safety of the technical system is not compromised. It is also crucial for classifying the device in a high safety category, such as SIL 2 and above.

[0052] It is understood that the features mentioned above and those to be explained below can be used not only in the combinations specified, but also in other combinations or on their own, without leaving the scope of the present invention.

[0053] Exemplary embodiments of the invention are shown in the drawing and are explained in more detail in the following description. Fig. Figure 1 shows, using the example of an industrial manufacturing cell, a possible application scenario for an embodiment of a device for the reliable detection of the absence of persons in a monitored area of ​​a technical plant. Fig. Figure 2 shows a schematic representation of an embodiment of a device for the reliable detection of the absence of persons in a monitored area. Fig. Figure 3 shows a schematic example of a deductive conclusion. Fig. Figure 4 shows a first variant for the implementation of agent tests in a device for the secure absence detection of persons. Fig. Figure 5 shows a second variant for implementing agent tests on a device for the reliable detection of the absence of persons.

[0054] Fig. Figure 1 shows, using an example of an industrial manufacturing cell, a possible application scenario for an embodiment of the device for the reliable detection of the absence of persons in a monitored area of ​​a technical plant.

[0055] In Fig. In Figure 1, the industrial production cell 12 is surrounded by a separating protective device. Here, the separating protective device is a protective enclosure 14, i.e., a complete enclosure that surrounds the entire machine or large parts of it and restricts access to specific, controlled points. Of course, the separating protective device could also be another physical barrier, such as a safety grille or fence, that prevents unauthorized access to the machine or other hazardous areas.

[0056] Within the protective enclosure 14, three industrial robots 16 are arranged as an example for a technical system, each mounted on a base 18. The industrial robots 16 are equipped with grippers 20 that pick up workpieces 22 from a conveyor belt 24 and transport them to a processing station 26.

[0057] The protective enclosure 14 is accessible from one side via a safety door 28, which provides access to the industrial production cell 12 inside the protective enclosure 14. Radar sensors 10 are installed to monitor the area within the protective enclosure 14 and safeguard the protected space defined by it. The radar sensors 10 are positioned at the corners and entrances of the protective enclosure 14 to ensure complete coverage of the robot's working area.

[0058] In the event of unauthorized entry of a person or object into the protective enclosure, the radar sensors 10 detect this intrusion and, by means of a safety switching device, bring the industrial robots 16 to a safe state. This safe state can be an immediate stop of movements, e.g., by switching off the power supply to the robots, or assuming a predefined safe position to prevent accidents or damage.

[0059] A control unit 30 of the production cell 12 is located outside the protective enclosure 14 and enables the operation and monitoring of the robots 16. The safety systems of the production cell 12 are connected to the safety switching device, e.g., a safety controller, which is either integrated into the control unit 30 or designed separately. In addition to the radar sensors 10 for monitoring the protective area, which act as sensors for the safety systems, the safety door 28 is also equipped with a safety switch 32 that allows the door to be opened only when the robots 16 are at a standstill.

[0060] The device for the reliable detection of persons' absence in a monitored area can be seamlessly integrated into the described scenario of the industrial manufacturing cell 12. In this application example, the radar sensors 10 are already present to monitor the area within the protective enclosure 14. These radar sensors 10 can function as sensors for the claimed device by continuously acquiring sensor data and forwarding it to the feature extraction unit.

[0061] The feature extraction unit analyzes the data acquired by the radar sensors 10 to extract human-specific features such as movement patterns, heartbeat, or respiration. The extracted features are then transmitted to the deductive decision module. The decision module deductively evaluates these features to determine whether the presence of a person can be ruled out.

[0062] Should the decision module conclude, based on the extracted features, that a person may be present in the monitored area, a safety-related action is initiated. In this scenario, this could mean activating the safety switching device to bring the industrial robots 16 into a safe state. This could involve immediately stopping the robots' movements or moving them to a safe position to ensure the safety of any person potentially present. It is also conceivable that another safety-related action is performed, such as switching the industrial robots' operating mode to one in which there is no longer any danger to people.

[0063] Integrating the claimed device improves the existing safety system of the production cell. The radar sensors 10 collect comprehensive data, while the feature extraction unit and decision module ensure that this data is reliably and securely evaluated. This reduces the probability of false alarms and increases overall safety, as the system can react dynamically to potential hazards. The safety switching device remains the central element for implementing safety measures, supported by the precise and reliable detection of the device. Advantageously, in addition to the radar sensors 10, further sensors (not shown here) can be provided to supply additional sensor data relating to the environment of the production cell, which can be evaluated with regard to human-specific characteristics.

[0064] Fig. Figure 2 shows a schematic representation of an embodiment of a device for the reliable detection of the absence of persons in a monitored area.

[0065] The device is here in its entirety designated by the reference numeral 100 and comprises three units: a data acquisition unit 102, a feature extraction unit 104 and a deductive decision module 106.

[0066] The units are shown here as individual components, but can also be designed as logical components of a common component. For example, the data acquisition unit 102, the feature extraction unit 104, and the deductive decision module can be implemented by an integrated circuit, such as a microcontroller or a system-on-a-chip. The units can therefore be arranged in a housing 105 and form a self-contained assembly. In another embodiment, the units can also be distributed across different devices as logical modules and interconnected via a communication medium. For example, it is conceivable to outsource the feature extraction unit to a high-performance computing unit, to which the sensor data is transmitted and which returns the extracted features.

[0067] The data acquisition unit 102 can be coupled with at least one sensor and serves to acquire sensor data from the environment. This sensor data can take various forms, such as visual data from cameras, thermal images, radar signals, or other sensor types suitable for detecting objects and their properties.

[0068] The feature extraction unit 104 is responsible for extracting human-specific features from the recorded sensor data. These features can include various physiological or behavioral indicators, such as the contours of the human body, movement patterns, body temperature, heartbeat, or other characteristic features that indicate the presence of a human.

[0069] The deductive decision module 106 uses the extracted features to determine, through deductive reasoning, whether the presence of a person in the monitored area cannot be ruled out. If, based on the detected features, there is a possibility that a person is in the monitored area, the module initiates a safety-related action. This action could include, for example, stopping or slowing down a machine, triggering an alarm, or other measures to ensure safety.

[0070] Deductive reasoning means drawing conclusions about specific cases from general principles or theories. It begins with a general statement or premise, which is assumed to be true, and derives specific, necessarily true conclusions from it. An example is the classic logical argument: "All humans are mortal. Socrates is a human. Therefore, Socrates is mortal." Here, a conclusion is drawn from a general rule (all humans are mortal) to a specific case (Socrates).

[0071] Inductive reasoning, on the other hand, proceeds from the specific to the general. It begins with specific observations or empirical data and derives general principles or theories from them. Inductive reasoning is probabilistic in nature, meaning it leads to conclusions that are likely, but not necessarily, true. An example would be observing several white swans and concluding that all swans are white. However, this conclusion is always vulnerable to new observations that could refute it, such as the appearance of a black swan.

[0072] For the deductive decision module 106 to reach a deductive conclusion, the choice of question it is to decide is crucial. In contrast to known devices, the question in this device is not whether a person is present, but rather whether the presence of a person cannot be ruled out based on the available sensor data. This decision can be made deductively. The deductive decision module can therefore be an inference machine.

[0073] As in Fig. As shown in Figure 2, the device can include further units. An input unit 108 and an output unit 110 are shown here as examples. The input unit 108 can be configured as one or more interfaces via which the device 100 can be coupled to one or more sensors 112 to acquire the sensor data. Various sensors are suitable as sensors 112, such as a camera 114 or a radar sensor 116. The sensors 112 can, in particular, be intrinsically safe sensors.

[0074] The output unit 110 can also be configured as one or more interfaces and connect the device to actuators (not shown here) to initiate the safety-related action when indicated by the deductive decision module. The safety-related action can also be implemented by a downstream unit, such as a control system for the technical plant or a safety controller. In this case, the output unit 110 can be a communication interface through which a conclusion from the deductive decision module 106 is transmitted for further processing.

[0075] Fig. Figure 3 shows a schematic representation of an example of a deductive conclusion.

[0076] The input side here indicates the individual human-specific characteristics 118 that can be provided by the feature extraction unit. In this example, the deductive decision module is configured to combine the characteristics with a logical OR 120 (ORing) and make a decision based on this. In other words, the deductive decision module makes a decision if one of the characteristics is present. The question 122, which is answered by the decision module, is whether the presence of a person cannot be ruled out. This question is answered with YES if at least one human-specific characteristic is displayed. The decision module thus operates purely rule-based and deterministic and is therefore verifiable and auditable in itself.This can be of great importance in a safety analysis and is advantageous compared to probabilistic approaches based on inductive reasoning.

[0077] Fig. 4 and Fig. Figure 5 shows variants for the implementation of agent tests in a device for the secure absence detection of persons.

[0078] In safe automation technology, "agent testing" refers to a specific testing method used to verify software or hardware components, known as agents, to ensure their reliable operation in safety-critical systems. These agents are often autonomous units that perform tasks within a larger system, and their correct behavior is crucial for the overall system's safety. Agent testing evaluates whether the agents respond correctly to various scenarios and inputs and whether they meet the defined safety requirements.

[0079] In the present device, an agent test is understood to be a method for verifying the functionality of algorithms or neural networks used in safety-critical applications. Specially prepared test data is fed into the system instead of the actual sensor data to ensure that the feature extraction unit and the associated algorithms are functioning correctly and delivering reliable results.

[0080] The agent test can, as in Fig. As shown in Figure 4, the tests are initiated by a safety controller that determines when and how they are performed. The test dataset is designed to cover a wide range of human-specific characteristics and simulate various scenarios that might occur in a real-world environment. During the test, the feature extraction unit processes this test data, and the decision module draws its conclusions based on this data.

[0081] If the device is integrated into a sensor and the sensor itself has a safety-related device 128, the agent test can also be implemented within the sensor, as in Fig. 5 shown.

[0082] It should be noted that the foregoing embodiments are only exemplary and further variations of individual components are possible to realize embodiments of the following claims. The scope of protection of the present invention is determined by the following claims and is not limited by the features explained in the description or illustrated in the figures.

Claims

[1] Device (100) for the reliable detection of the absence of persons in a monitored area of ​​a technical installation, comprising: a data acquisition unit (102) which is configured to record sensor data relating to an environment of the technical installation, provided by at least one sensor, a feature extraction unit (104) which is submitted to extract human-specific features from the sensor data, as well as a deductive decision module (106) that is set up to initiate a security-related action when, based on a deductive inference derived from the extracted human-specific characteristics, the presence of a person cannot be ruled out. [2] Device according to claim 1, wherein the deductive decision module (106) is an inference machine. [3] Device according to claim 1 or 2, wherein the deductive inference involves linking the extracted human-specific features with a logical OR. [4] Device according to one of claims 1 to 3, wherein the feature extraction unit (104) is configured to extract and analyze several human-specific features simultaneously. [5] Device according to any one of claims 1 to 4, wherein the feature extraction unit (104) is a KL-based feature extraction unit which uses a trained neural network for feature extraction. [6] Device according to any one of claims 1 to 5, wherein the sensor data are data from a camera (114), in particular a 3D camera or a thermal imaging camera. [7] Device according to any one of claims 1 to 6, wherein the sensor data are data from a radar sensor (116). [8] Device according to any one of claims 1 to 7, wherein the feature extraction unit (104) is configured for the detection of contours, skeleton, heartbeat, respiration, temperature, energy density, and / or movement patterns. [9] Device according to any one of claims 1 to 8, wherein the safety-related action includes an immediate shutdown of the technical system or a transfer of the technical system to a safe state. [10] Device according to any one of claims 1 to 9, wherein the sensor data are data from an intrinsically safe sensor. [11] Device according to one of claims 1 to 10, wherein the feature extraction unit (104) is not implemented in a safety-oriented manner. [12] Device according to any one of claims 1 to 11, wherein the technical system is operable in a first mode in which the technical system poses a danger to persons and objects in the vicinity of the technical system, and is operable in a second mode in which the technical system poses no danger or a manageable danger to persons and objects in the vicinity of the technical system, and wherein the safety-related action comprises switching from the first mode to the second mode if the presence of a person in the vicinity cannot be ruled out. [13] Device according to claim 12, wherein the device has a further safety device which is activated in the second mode to ensure end protection. [14] Device according to any one of claims 1 to 13, further comprising: a test unit configured to verify the functionality of the feature extraction unit (104) by means of an agent test. [15] Method for the reliable detection of the absence of persons in a monitored area of ​​a technical installation, comprising: - Recording sensor data related to an environment of the technical system with at least one sensor; - Extracting human-specific characteristics from sensor data using a feature extraction unit; - Initiating a security-related action through a deductive decision module if, based on a deductive conclusion derived from the extracted human-specific characteristics, the presence of a person cannot be ruled out.

Images