Method and device for classifying the behavior of a pedestrian when crossing a vehicle's roadway, as well as a vehicle's personal protection system.

Abstract

Method (300) for classifying the behavior of a pedestrian (115) when crossing a roadway (110) of a vehicle (100), wherein the method (300) comprises the following steps: Reading (305) a sensor signal to detect the pedestrian (115) and at least one environmental information (200, 220) relating to the environment of the pedestrian (115), wherein the sensor signal represents a signal from at least one sensor (120) of the vehicle (100), and wherein at least one pedestrian (220) crossing the roadway (110) is detected as the environmental information; Determine (310) at least one physical quantity (d, t, v) of a relationship between the pedestrian (115) and the at least one piece of environmental information (200, 220), wherein a distance (d) between the pedestrian (115) and the pedestrian (220) crossing the roadway (110) is determined as a physical quantity; and Classifying (315) the behavior of the pedestrian (115) using at least one physical quantity (d, t, v), wherein the behavior of the pedestrian (115) is classified using the distance (d).

Description

State of the art

[0001] The present invention relates to a method for classifying the behavior of a pedestrian when crossing a roadway of a vehicle, to a corresponding device, to a corresponding computer program product, to a corresponding storage medium and to a personal protection system.

[0002] Implementing active pedestrian protection systems generally requires predicting pedestrian movement, i.e., estimating the possible locations of a pedestrian at risk at future times. Based on such a prediction, a decision can be made as to whether to initiate emergency braking or emergency evasive action to avoid an imminent accident.

[0003] US 2010 / 0076621A1 discloses a device capable of detecting a dangerous moving object with high accuracy and fast timing.

[0004] DE 10 2008 062 916 A1 discloses a method for determining the probability of a collision between a vehicle and a living being, in which the spatiotemporal behavior of the living being is modeled with a behavioral model and the spatiotemporal behavior of the vehicle is modeled with a kinematic model, and at least one trajectory is determined starting from the current positions of the vehicle and the living being.

[0005] US 2005 / 0073438A1 discloses a method for providing a driver with warnings about potential vehicle-pedestrian collisions. Disclosure of the invention

[0006] Against this background, the approach described here presents a method and a device for classifying the behavior of a pedestrian when crossing a vehicle's lane, a corresponding computer program product, a corresponding storage medium, and a pedestrian protection system according to the main claims. Advantageous embodiments are described in the respective dependent claims and the following description.

[0007] The present approach creates a procedure for classifying the behavior of a pedestrian when crossing a roadway of a vehicle, the procedure comprising the following steps: Reading a sensor signal to detect the pedestrian and at least some environmental information relating to the pedestrian's environment, wherein the sensor signal represents a signal from at least one sensor of the vehicle;

[0008] Determine at least one physical quantity relating to a relationship between the pedestrian and at least one piece of environmental information; and

[0009] Classifying pedestrian behavior using at least one physical quantity.

[0010] A pedestrian is defined as a person who is in the path of a vehicle. Based on the pedestrian's behavior, inferences can be drawn about their possible movement or direction of movement within the path. A pedestrian's behavior can include, for example, their level of attention or willingness to cooperate. A path can be understood as, for example, a single- or multi-lane road used by a vehicle. A vehicle can be understood as, in particular, a motor vehicle such as a car, truck, or motorcycle. The pedestrian and their surroundings can be detected by a sensor on the vehicle. The pedestrian's surroundings can be understood as an area of ​​the path that is at least partially visible or accessible to the pedestrian or the driver of the vehicle.The surroundings can include not only the roadway but also objects such as parked and moving vehicles or other obstacles, for example, trees or traffic islands. The term "surroundings" is not limited to the roadway; it can also encompass all objects and other elements such as lines and curbs. Features or elements of the surroundings detected by a sensor can be represented by environmental information. This environmental information can be represented by an electrical signal or machine-readable data. The environmental information can then be further processed by a suitable device. For example, the surroundings could be the area in front of the vehicle, within which a pedestrian might be located. A sensor could be, for example, a camera on the vehicle pointed at this area in front of the vehicle.Environmental information can be understood as information about the surroundings that is relevant for assessing a pedestrian's behavior. For example, environmental information can represent an obstacle, an open space, or other objects such as another pedestrian or vehicle in the pedestrian's vicinity, or even the position and course of curbs and road markings. A relationship can exist between the pedestrian and the environmental information. This relationship can be understood as a spatial, temporal, or spatiotemporal connection that can be expressed by a corresponding physical quantity. A physical quantity could, for example, be a measured value of speed, acceleration, distance, or time period.

[0011] The present approach is based on the understanding that information from a pedestrian's environment can be used to determine a pedestrian's possible behavior when crossing a roadway.

[0012] For example, a reliable prediction of a possible pedestrian movement can be made based on such determined behavior.

[0013] Such context-dependent prediction can also be carried out using features that are directly assigned to the pedestrian.

[0014] The information or characteristics can be determined, for example, from data from a vehicle's video environmental sensor system, in particular a forward-facing stereo video camera.

[0015] This allows the benefit of prediction to be increased compared to conventional methods and a false trigger rate to be reduced, especially in the case of an emergency evasive maneuver by the vehicle.

[0016] To avoid false alarms, emergency braking systems can use a method that predicts all possible trajectories a pedestrian could physically take. Emergency braking is only triggered if all possible trajectories, or a very large proportion (for example, 90%), would lead to a collision.

[0017] For systems designed to avoid collisions with pedestrians through an evasive maneuver, possibly combined with automatic emergency braking, a maximum assumption may be unsuitable, as it can lead to an overestimation of the actual required evasive clearance or fail to provide a clear decision regarding the direction and width of the evasive maneuver. The pedestrian could potentially be in many different locations in the future. To plan an appropriate evasive maneuver nonetheless, more conservative assumptions regarding future locations can be used. For example, it can be assumed that the pedestrian continues moving at a constant speed. However, this is not appropriate in all situations. For instance, the pedestrian often stops just before crossing the road because they have already seen an approaching vehicle. To account for this, a possible pedestrian stop line can be determined.For example, movement characteristics can be used, such as an optical flow in the leg and upper body region of the pedestrian, to predict whether the pedestrian will stop or walk across the street.

[0018] According to one embodiment of the present approach, the method can include a step of determining at least one possible trajectory of the pedestrian, depending on the pedestrian's behavior classified in the classification step. A possible trajectory can be understood as a possible movement pattern of the pedestrian when crossing the roadway. By determining at least one possible trajectory of the pedestrian, a potential collision between the pedestrian and the vehicle can be detected at an early stage.

[0019] According to a further embodiment of the present approach, the method can include a step of providing an activation signal to activate a vehicle safety device depending on the possible trajectory. A safety device can be, for example, an occupant protection device, such as one or more airbags, or a pedestrian protection device, such as an automatic braking or evasive maneuver by the vehicle. This embodiment of the present approach can reduce the risk of injury to pedestrians, vehicle occupants, or other road users.

[0020] Furthermore, during the data acquisition step, an environmental model of the pedestrian's surroundings can be created using the sensor signal to capture the pedestrian and environmental information. An environmental model can be understood as a database for the detailed and real-time representation of the pedestrian's environment. Using such an environmental model, the pedestrian and environmental information can be captured very precisely and reliably.

[0021] Furthermore, in the classification step, pedestrian behavior can be classified using a hidden Markov model and / or a support vector machine and / or fuzzy logic and / or neural networks. This allows for a robust estimation of pedestrian behavior.

[0022] Furthermore, during the data acquisition step, at least one potential pedestrian stop line can be captured as environmental information. In the determination step, the speed of relative movement between the pedestrian and the potential stop line can be determined. In the classification step, the pedestrian's behavior can be classified using this speed of relative movement. A potential stop line can be understood as an actual or hypothetical line at which the pedestrian stops before crossing the roadway. This potential stop line could be, for example, a road marking or the edge of the road. Using the speed of relative movement between the pedestrian and the potential stop line, it is possible to estimate whether, and at what point, the pedestrian will cross the roadway.

[0023] According to a further embodiment of the present approach, at least one pedestrian crossing the roadway can be captured as environmental information during the input step. In the detection step, a distance between the pedestrian and the pedestrian crossing the roadway can be determined. During the classification step, the pedestrian's behavior can be classified using this distance. The distance allows for an estimation of whether the pedestrian is directly following the pedestrian crossing the roadway or not.

[0024] Furthermore, in the input step, at least one possible crossing point for the pedestrian can be captured as environmental information. In the determination step, the distance between the pedestrian and the possible crossing point can be calculated. In the classification step, the pedestrian's behavior can be classified using this distance. A possible crossing point could be, for example, a bus stop, a building, a crosswalk, or a traffic light. This embodiment of the present approach also allows for a highly accurate determination of the probability that the pedestrian will cross the roadway.

[0025] According to a further embodiment of the present approach, the crossing location can also be determined using GPS information and / or map information during the detection step. This allows for high accuracy in determining the distance between the pedestrian and the potential crossing point.

[0026] Furthermore, during the data acquisition step, a possible line of sight between the vehicle and the pedestrian can be captured as environmental information. In the determination step, a time span of this possible line of sight can be determined. During the classification step, the pedestrian's behavior can be classified using this time span. A possible line of sight can be understood as a clear area between the vehicle and the pedestrian. This line of sight implies mutual visibility between the pedestrian and the vehicle driver. The time span can thus be understood as the duration during which the pedestrian could have seen the vehicle, taking into account any obstructions. The longer this time span, the more likely it is that the vehicle was perceived by the pedestrian. Using this embodiment of the present approach, the pedestrian's level of attention can be assessed.

[0027] During the data acquisition step, the pedestrian's height and / or gaze direction can also be recorded. In the classification step, the pedestrian's behavior can then be further classified based on this height and / or gaze direction. This embodiment of the present approach also allows for a reliable assessment of the pedestrian's attention.

[0028] The approach presented here further provides a device designed to carry out or implement the steps of a variant of the method presented herein in appropriate facilities. This embodiment of the invention in the form of a device also allows the problem underlying the invention to be solved quickly and efficiently.

[0029] In this context, a device can be understood as an electrical device that processes sensor signals and outputs control and / or data signals accordingly. The device may have an interface, which can be implemented in hardware and / or software. In the case of a hardware-based interface, the interfaces can, for example, be part of a so-called system ASIC, which incorporates various functions of the device. However, it is also possible that the interfaces are separate integrated circuits or consist at least partially of discrete components. In the case of a software-based interface, the interfaces can be software modules, which, for example, are present on a microcontroller alongside other software modules.

[0030] It is also advantageous to have a computer program product with program code that can be stored on a machine-readable medium such as semiconductor memory, hard disk memory or optical memory and is used to carry out the method according to one of the embodiments described above, if the program product is executed on a computer or device.

[0031] Furthermore, the present approach creates a machine-readable storage medium with a computer program stored on it according to a previously described embodiment.

[0032] Finally, the present approach creates a personal protection system for a vehicle, the personal protection system having the following features: at least one sensor for detecting a pedestrian and environmental information regarding the pedestrian's surroundings; a device connected to at least one sensor according to a previously described embodiment; and a personal protective device that is designed to be activated by the device.

[0033] The approach presented here is explained in more detail below using the attached drawings as examples. These show: Fig. 1 a schematic representation of a vehicle with a personal protection system according to an embodiment of the present invention; Fig. 2a, Fig. 2b Schematic representations of a stop line for use in a method according to an embodiment of the present invention; Fig. 3 a flowchart of an embodiment of a method for classifying the behavior of a pedestrian when crossing a roadway of a vehicle; Fig. 4 a system structure of a personal protection system according to an embodiment of the present invention; and Fig. 5 a block diagram of an embodiment of a device for classifying the behavior of a pedestrian when crossing a vehicle's roadway.

[0034] In the following description of favorable embodiments of the present invention, the same or similar reference numerals are used for the elements shown in the various figures and acting similarly, without repeating these elements.

[0035] Fig. Figure 1 shows a schematic representation of a vehicle 100 with a pedestrian protection system 105 according to an embodiment of the present invention. The vehicle 100 is located on a roadway 110. A pedestrian 115 is located in front of the vehicle 100. The pedestrian 115 is about to cross the roadway 110. The pedestrian protection system 105 comprises a sensor 120, a pedestrian protection device 125, and a control unit 130. The control unit 130 is connected to the sensor 120 and the pedestrian protection device 125. The sensor 120 is arranged in a front area of ​​the vehicle 100 and is directed towards the area in front of the vehicle 100.

[0036] Sensor 120 is configured to detect pedestrian 115 and its surroundings. Control unit 130 is configured to receive a sensor signal from sensor 120 representing the surroundings and pedestrian 115, and to use this sensor signal to determine a physical quantity relating to the relationship between the surroundings and pedestrian 115. Furthermore, control unit 130 is configured to classify the behavior of pedestrian 115 based on this physical quantity. An example is provided in Fig. 1 the behavior of pedestrian 115 is classified by the control unit 130 as the behavior of a pedestrian crossing the roadway 110.

[0037] The control unit 130 is additionally configured to determine at least one possible trajectory 135 of the pedestrian 115 when crossing the roadway 110, depending on the pedestrian's behavior. The control unit 130 can be configured to send an activation signal to the pedestrian protection device 125, depending on the possible trajectory 135.

[0038] The personal protection device 125 is designed, for example, to cause an evasive movement 140 of the vehicle 100 in response to the reception of the activation signal.

[0039] Thus, the vehicle 100 can avoid the pedestrian 115 crossing the roadway 110 in time to prevent a collision between the vehicle 100 and the pedestrian 115.

[0040] Fig. 2a, Fig. Figure 2b shows schematic representations of a stop line 200 for use in a method according to an embodiment of the present invention.

[0041] Fig. Figure 2a shows lane 110 with vehicle 100. Two other vehicles 205 are parked at one side of lane 110. The other vehicles 205 are arranged one behind the other. The pedestrian 115 is moving towards the stop line 200 in a gap between the other vehicles 205. Fig. 2a is not a directly visible stop line, but a fictitious stop line 200. The stop line 200 results from a connecting line between the parked cars 205.

[0042] The vehicle 100 is designed to detect the stop line 200 and the pedestrian 115 and to determine a physical quantity relating to the relationship between the stop line 200 and the pedestrian 115. For example, the vehicle 100 is designed to determine the speed v of any relative movement between the stop line 200 and the pedestrian 115 and, depending on the speed v, to classify the behavior of the pedestrian 115 with regard to a possible movement pattern when crossing the roadway 110.

[0043] Unlike Fig. 2a corresponds to stop line 200 in Fig. 2b shows, for example, a lane marking 210 at the edge of the roadway. For example, a median strip 215 or other markings of the roadway 110 can also be recorded as stop lines 200.

[0044] In Fig. Figure 2b shows an additional pedestrian 220. Unlike pedestrian 115, who is at the edge of the roadway, pedestrian 220 has already crossed most of the roadway 110. According to one embodiment of the present invention, the vehicle 100 is designed to further detect pedestrian 220 and determine a distance d between pedestrian 115 and pedestrian 220. Depending on the value of this distance d, a possible behavior of pedestrian 115 can be inferred. For example, the behavior of pedestrian 115 can be classified as that of a pedestrian crossing the roadway 110 if the distance d is relatively small, for instance, compared to the width of the roadway, which is also detected.

[0045] According to a further embodiment of the present invention, the vehicle 100 is configured to detect a possible line of sight between the vehicle 100 and the pedestrian 115 and to determine a time interval t of the possible line of sight. Alternatively or additionally, the behavior of the pedestrian 115 can be classified as a function of the time interval t.

[0046] Fig. Figure 3 shows a flowchart of an embodiment of method 300 for classifying the behavior of a pedestrian crossing a vehicle's lane. Method 300 comprises a step 305 of reading a sensor signal to detect the pedestrian and at least one piece of environmental information relating to the pedestrian's environment, wherein the sensor signal represents a signal from at least one sensor of the vehicle. Method 300 further comprises a step 310 of determining a physical quantity relating to the relationship between the pedestrian and the at least one piece of environmental information. Finally, method 300 comprises a step 315 of classifying the pedestrian's behavior using the at least one physical quantity.

[0047] According to one embodiment of the present invention, in step 305 of the input process, at least one possible crossing point of the pedestrian is captured as environmental information. In step 310 of the determination process, a distance between the pedestrian and the possible crossing point is determined as a physical quantity. Finally, in step 315, the behavior of the pedestrian is classified using this distance.

[0048] Fig. Figure 4 shows a system structure 400 of a personal protection system according to an embodiment of the present invention. First, in step 405, sensor data is preprocessed. This data is provided, for example, by a stereo video sensor directed at the area in front of a vehicle. During preprocessing, a disparity map or classifiers are created, for example. Based on step 405 of the preprocessing, an environment model of the vehicle's surroundings is created in step 410. The environment model can include, for example, an occupancy map, an object list, or a description of open space.

[0049] Using the sensor data preprocessed in step 405 and / or the environment model created in step 410, a feature determination is performed in step 415. This involves, for example, determining the relationship of a pedestrian to a fictitious stop line or other pedestrians, or the duration of mutual visibility between the vehicle and the pedestrian.

[0050] Responding to step 415 of the feature determination, step 420 of the behavior classification takes place. For example, the pedestrian's behavior is classified as "stops" as opposed to "crosses", as "attentive" as opposed to "inattentive", or as "cooperative" as opposed to "uncooperative".

[0051] Finally, depending on the behavior classification carried out in step 420, step 425 involves a prediction based on the classified behavior.

[0052] According to a report in the Fig. 1, Fig. 2a, Fig. 2b and Fig. In the embodiment of the present invention shown in Figure 4, an environment model is estimated based on environmental sensors 120 and corresponding raw data preprocessing, such as determining a disparity map from stereo video sensors or classifying the grayscale values ​​of a video image. This model contains information about obstacles, for example in the form of an obstacle map, an occupancy grid, a free area, or in the form of objects.

[0053] Based on this data, features are extracted that are used to classify the behavior of pedestrian 115. Possible classes of pedestrian behavior include, for example, "pedestrian 115 stops" as opposed to "pedestrian 115 crosses the road 110", "pedestrian 115 changes direction" as opposed to "pedestrian 115 maintains direction", "pedestrian 115 has seen the ego-vehicle 100" as opposed to "pedestrian 115 has not seen the ego-vehicle 110", or "pedestrian 115 behaves cooperatively," i.e., helps to avoid an accident, as opposed to "pedestrian 115 does not behave cooperatively".

[0054] The classification of pedestrian behavior based on features can be implemented using methods such as Hidden Markov Models (HMMs), where each hidden state corresponds to a pedestrian behavior, Support Vector Machines (SVMs), fuzzy logic, or neural networks (NNs). These methods allow for the fusion of multiple features, enabling a robust estimation of pedestrian behavior.

[0055] Based on the detected pedestrian behavior, a suitable prediction can then be made. For example, such a context-dependent prediction of pedestrian 115 can be used to control an active pedestrian protection system 105.

[0056] The following describes several characteristics that can be used together to classify pedestrian behavior. It is also possible to use only a subset of these characteristics.

[0057] An important characteristic for classifying pedestrian movement is the relative movement of the pedestrian 115 to a potentially fictitious stop line 200. This is based on the model that the pedestrian 115, who wants to cross the road 110, imagines a line at which he will stop if traffic does not allow him to cross immediately. Such stop lines 200 are realized, for example, as roadway boundaries, such as a curb, lane markings 210, 215, or boundaries to an area that the vehicle 100 is unlikely to enter, such as an area between two parked vehicles 205, as in Fig. 2a shown.

[0058] A positive or negative acceleration of the pedestrian 115 can be used as a characteristic, which is necessary for the pedestrian 115 to come to a stop before the stop line 200. If the acceleration is very high, this can be seen as an indication of a pedestrian state "crossing the road 110", "has overlooked the ego-vehicle 100", or "is not cooperating".

[0059] Another feature is the movement of other pedestrians 220. If pedestrian 220 has already crossed street 110, this can be used as an indication that pedestrian 115 standing at the roadside is following the preceding pedestrian 220 and will also cross street 110. A distance d to the already crossing pedestrian 220 can be used as a feature here.

[0060] The size of a pedestrian (115) can also be used as an indicator of their behavior. A small pedestrian (115) might be a child, who is more likely to exhibit the states "crossing the street 110," "overlooked the ego-vehicle 100," or "uncooperative" than a large pedestrian (115), who might be an adult.

[0061] The longer pedestrian 115 and the ego-vehicle 100 are mutually visible, i.e., not obscured from pedestrian 115's ego-perspective, the higher the probability that pedestrian 115 will "stop," "see the ego-vehicle 100," or "behave cooperatively." Conversely, brief mutual visibility suggests the other behavioral states.

[0062] The immediate proximity of pedestrian 115 to specific locations such as a bus stop, school, or kindergarten, determined alternatively or additionally using GPS or map data, increases the probability of the states "crossing the street 110," "overlooked the ego-vehicle 100," or "not cooperating." In the case of a bus stop, the presence of a stopped bus at the stop can also be detected.

[0063] Zebra crossings, traffic lights, or road signs increase the probability of the state "pedestrian 115 crosses road 110". To clearly determine alternative pairs of states such as "pedestrian 115 has seen the ego-vehicle 100" as opposed to "pedestrian 115 has not seen the ego-vehicle 100" and "pedestrian 115 is cooperative" as opposed to "pedestrian 115 is uncooperative", this feature can, for example, be compared with other features mentioned.

[0064] Optionally, the pedestrian's gaze direction can be used to determine whether the pedestrian 115 has already seen the approaching vehicle 100 or not. The gaze direction can also be used to determine what destination the pedestrian 115 is aiming for or whether he plans to change his direction of movement.

[0065] Fig. Figure 5 shows a block diagram of an embodiment of a device 500 for classifying the behavior of a pedestrian when crossing a vehicle's lane. The device 500 can be the one described in Fig.The device 500, also called a pedestrian movement classifier, comprises a unit 505 for reading a sensor signal to detect the pedestrian and at least one piece of environmental information relating to the pedestrian's environment, wherein the sensor signal represents a signal from at least one sensor of the vehicle. The unit 505 is connected to a unit 510 for determining at least one physical quantity relating to the relationship between the pedestrian and the at least one piece of environmental information. Finally, the device 500 comprises a unit 515, which is connected to the unit 510 and is configured to classify the pedestrian's behavior using the at least one physical quantity.

[0066] The embodiments described and shown in the figures are only examples. Different embodiments can be combined completely or with respect to individual features. An embodiment can also be supplemented by features from another embodiment.

[0067] Furthermore, the procedural steps presented here can be repeated and carried out in a different order than described.

[0068] If an embodiment includes an “and / or” connection between a first feature and a second feature, this is to be read as meaning that the embodiment according to one embodiment has both the first feature and the second feature, and according to another embodiment either only the first feature or only the second feature.

Claims

[1] Method (300) for classifying the behavior of a pedestrian (115) when crossing a roadway (110) of a vehicle (100), wherein the method (300) comprises the following steps: Reading (305) a sensor signal to detect the pedestrian (115) and at least one environmental information (200, 220) relating to the environment of the pedestrian (115), wherein the sensor signal represents a signal from at least one sensor (120) of the vehicle (100), and wherein at least one pedestrian (220) crossing the roadway (110) is detected as the environmental information; Determine (310) at least one physical quantity (d, t, v) of a relationship between the pedestrian (115) and the at least one piece of environmental information (200, 220), wherein a distance (d) between the pedestrian (115) and the pedestrian (220) crossing the roadway (110) is determined as a physical quantity; and Classifying (315) the behavior of the pedestrian (115) using at least one physical quantity (d, t, v), wherein the behavior of the pedestrian (115) is classified using the distance (d). [2] Method (300) according to claim 1, comprising a step of determining at least one possible trajectory (135) of the pedestrian (115) depending on the behavior of the pedestrian (115) classified in the classifying step (315). [3] Method (300) according to claim 2, comprising a step of providing an activation signal to activate a personal protection device (125) of the vehicle (100) depending on the possible trajectory (135). [4] Method (300) according to one of the preceding claims, wherein in step (305) of reading, an environment model of the environment of the pedestrian (115) is created using the sensor signal in order to detect the pedestrian (115) and the environment information (200, 220). [5] Method (300) according to one of the preceding claims, wherein in step (305) of reading at least one possible stop line (200) of the pedestrian (115) is recorded as environment information, wherein in step (310) of determining a speed (v) of a relative movement between the pedestrian (115) and the possible stop line (200) is determined as a physical quantity, and wherein in step (315) of classifying the behavior of the pedestrian (115) is classified using the speed (v) of the relative movement. [6] Method (300) according to one of the preceding claims, wherein in step (305) of reading at least one possible crossing location of the pedestrian (115) is recorded as environment information, wherein in step (310) of determining a distance between the pedestrian (115) and the possible crossing location is determined as a physical quantity and wherein in step (315) of classifying the behavior of the pedestrian (115) is classified using the distance. [7] Method (300) according to one of the preceding claims, wherein in step (305) of reading a possible line of sight between the vehicle (100) and the pedestrian (115) is recorded as environment information, wherein in step (310) of determining a time span (t) of the possible line of sight is determined as a physical quantity and wherein in step (315) of classifying the behavior of the pedestrian (115) is classified using the time span (t). [8] Method (300) according to one of the preceding claims, wherein in step (305) of reading a body size and / or a gaze direction of the pedestrian (115) is further recorded, wherein in step (315) of classifying the behavior of the pedestrian (115) is further classified depending on the body size and / or the gaze direction. [9] Device (130, 500) configured to perform all steps of a method (300) according to claim 1. [10] Computer program configured to perform all steps of a method (300) according to claim 1. [11] Machine-readable storage medium with a computer program stored thereon according to claim 10. [12] Personal protection system (105) of a vehicle (100), wherein the personal protection system (105) has the following features: at least one sensor (120) for detecting a pedestrian (115) and environmental information (200, 220) regarding the pedestrian's environment (115); a device (130, 500) connected to the at least one sensor (120) according to claim 9; and a personal protection device (125) designed to be activated by the device (130, 500).

Images