Method for classifying an occupant on a vehicle seat

Abstract

The invention relates to a method for classifying an occupant (1) on a vehicle seat in a vehicle and for identifying a seating position of the occupant (1) on the vehicle seat, wherein it is detected that the occupant (1) has fastened a safety belt (2) allocated to said occupant, and the body weight of the occupant (1) is determined. According to the invention, - at least one belt strap extension length of a belt strap extension identification system and captured image data of at least one vehicle interior camera are provided as input variables (E1 to En) to at least one prediction model (P) trained by means of artificial intelligence and - three-dimensional proportions of the occupant (1) and / or a course of the safety belt (2) on the body of the occupant (1) and / or a three-dimensional seating position of the occupant (1) on the vehicle seat are / is ascertained by means of the at least one prediction model (P).

Description

[0001] Mercedes-Benz Group AG

[0002] Procedure for classifying an occupant in a vehicle seat

[0003] The invention relates to a method for classifying an occupant on a vehicle seat in a vehicle and for recognizing the occupant's seating position on the vehicle seat, wherein it is detected that the occupant has fastened a seat belt assigned to him / her and the occupant's body weight is determined.

[0004] From DE 10 2019 004 864 A1 (D1) a method for estimating a driver's weight and height is known. With the driver's hands positioned on the steering wheel, the driver's seat is adjusted until the driver assumes a defined posture. Based on the seatbelt extension length and the driver's seat adjustment, the driver's weight and height are then estimated. This method thus uses the seatbelt extension length as one of several geometric parameters for the static determination of occupant characteristics.

[0005] The invention is based on the objective of providing a method for classifying an occupant on a vehicle seat in a vehicle, which enables improved and more reliable detection of dynamic, safety-relevant changes in the belt path.

[0006] The problem is solved according to the invention by a method which has the features specified in claim 1.

[0007] Advantageous embodiments of the invention are the subject of the dependent claims.

[0008] A method for classifying an occupant on a vehicle seat in a vehicle and for

[0009] The invention provides that the detection of an occupant's seating position on the vehicle seat, wherein it is detected that the occupant has fastened an assigned seat belt and the occupant's body weight is determined, is carried out by means of the following:

[0010] - at least a seatbelt extension length of a seatbelt extension detection system and captured image data from at least a vehicle interior camera are fed as input variables to a prediction model trained using at least one artificial intelligence system and

[0011] - using at least one prediction model, three-dimensional proportions of the occupant and / or a course of the seat belt on the occupant's body and / or a three-dimensional seating position of the occupant on the vehicle seat are determined, and a change in the course of the seat belt on the occupant's body is determined based on a recorded speed of a change in the determined belt extension length.

[0012] By applying this method, it is possible to classify an occupant and determine their seating position on their vehicle seat for all occupants in the vehicle. Furthermore, it enables the detection of any changes in the seating position of a given occupant during vehicle operation, as well as any changes in the position of the seatbelt on the occupant's body. In particular, by analyzing the speed of changes in belt length, dynamic manipulations of the belt, such as briefly lifting the belt from the upper body, can be distinguished from normal, slow changes in position, which significantly increases the reliability of the detection.

[0013] Regarding the occupant's seating position in the vehicle seat, this method makes it possible to determine, in particular, whether the occupant is sitting straight, at an angle, upright, flat, or hunched over. It can also identify whether the occupant is a so-called "long-legged" or "short-legged" person.

[0014] One result of a comprehensive validation using the predictive model is an optimized occupant and seat position classification for the optimized tuning and adjustment of the vehicle's restraint systems, the ignition timing of restraint systems, ignition logic, and / or information and error messages sent to an occupant to correct their seat position, particularly regarding the seatbelt's position on the occupant's body. All necessary components for this process are essentially already present in the vehicle, so no design modifications are required. For example, the predictive model can be implemented in a control unit for the vehicle's restraint systems.

[0015] In one implementation, at least one predictive model is trained using data from a conducted occupant study and / or image data of current vehicle interiors and / or data from an occupant simulation. Thus, the predictive model is trained to determine, based on the input variables, the three-dimensional proportions of the occupant, the path of the seatbelt on the occupant's body, and the three-dimensional seating position of the occupant in their vehicle seat, particularly in relation to the vehicle's restraint systems.

[0016] In one possible implementation, the seatbelt extension length is determined based on signals from at least one sensor used for belt extension detection. This provides an absolute value for the belt extension length, which is a crucial input for determining the belt's position when fastened, particularly to verify that the belt is correctly fastened. Furthermore, the belt extension length, combined with captured image data of the occupant, can be used to infer the occupant's seating position.

[0017] In a further embodiment of the method, a change in the seat belt's position on the occupant's body is detected based on a measured speed, a change in the determined belt extension length, and / or the changing belt extension length itself. This allows the system to recognize that the seat belt's position on the occupant's body has changed, for example, because the occupant has changed their seating position in the vehicle seat or because the occupant has unfastened the seat belt.

[0018] In another possible implementation, forward movement of the occupant in the direction of travel is detected using captured three-dimensional image data, and the resulting slippage of the seat belt from one of the occupant's shoulders is determined by the decreasing belt extension length. This means that the seat belt may provide insufficient restraint in an emergency, particularly in the event of a collision, which can result in an increased risk of injury for the occupant.

[0019] Furthermore, one implementation stipulates that a vehicle state and / or an activation state of at least one assistance system is determined and fed into the predictive model as input variables. Particularly in critical driving situations, it is crucial that the seat belt is fastened and the occupant maintains an upright seating position in their vehicle seat, so that optimized restraint is ensured as far as possible by means of the fastened seat belt.

[0020] In one implementation, the buckle status of the occupant's assigned seatbelt is recorded and fed into the predictive model, allowing it to be assumed that the seatbelt is properly fastened. Subsequently, the seatbelt's path along the occupant's body is determined, particularly using captured image data.

[0021] In a further implementation, if the predictive model determines that the fastened seat belt provides insufficient restraint for the occupant in the event of an accident, based on the occupant's classification and / or its position on the occupant's body, a warning message is displayed in the vehicle. This warning message advises the vehicle occupant, particularly the driver, to adjust their seating position to ensure the highest possible level of restraint provided by the seat belt, thereby reducing the risk of injury.

[0022] Exemplary embodiments of the invention are explained in more detail below with reference to drawings.

[0023] This shows:

[0024] Fig. 1 schematically shows an overview of different seating positions of an occupant on a

[0025] Vehicle seat in a vehicle, Fig. 2 schematically shows an occupant of a vehicle in different positions.

[0026] Head and

[0027] Fig. 3 schematically shows an overview of various input variables for a system in a

[0028] Control unit with stored trained predictive model for classifying the occupant and recognizing the occupant's seating position.

[0029] Corresponding parts are marked with the same reference symbols in all figures.

[0030] Figure 1 shows an overview of different seating positions of an occupant 1 on a vehicle seat (not shown in detail) in a vehicle.

[0031] Figure 2 shows an occupant 1 of a vehicle with different positions of his head 1.1 and in Figure 3 is an overview of different input variables El to En for a trained prediction model P stored in a control unit 3 for classifying the occupant 1 and for recognizing a seating position of the occupant 1 on a vehicle seat.

[0032] It is generally known that vehicle interior cameras capable of capturing two-dimensional and three-dimensional image data are able to determine a reference point B at the head 1.1 of the occupant 1 with respect to a vehicle coordinate system, also known as the vehicle axis system. Determination using three-dimensional image data is more accurate, whereby image data captured by a 3D vehicle interior camera is used to detect the distance between the head 1.1 of the occupant 1 and a steering wheel and / or to detect a relaxed position of the occupant 1 on the vehicle seat.

[0033] It is also possible to assign a determined position of the head 1.1 of occupant 1 relative to a position of the head 1.1 determined in a measurement concept and to derive a height of occupant 1. The tolerance range for classifying height is comparatively large, so that an unsuitable seating position of occupant 1 on their vehicle seat, for use with restraint systems, may not be detected or may only be detected at the limit of the tolerance range.

[0034] For this purpose, three-dimensional information about the seating position of occupant 1, holistically and depending on different proportions and postures, for example upright or curved, is important information which is required to classify occupant 1 and, if necessary, to adapt the vehicle's restraint systems in their triggering characteristics and triggering time.

[0035] The following describes a method that enables a holistic classification of an occupant 1 and their respective seating position on all seats in a vehicle. This method also allows for the detection of potential changes in the position of a seatbelt 2 fastened by an occupant 1 during vehicle operation. Depending on the determined seating position and the position of the seatbelt 2 on the occupant 1's body, a warning message is selected from a portfolio of possible errors and displayed in the vehicle.

[0036] The procedure stipulates that all input variables El to En, explained below, are combined and comprehensively validated. This validation results in the best possible classification of occupant 1 and their seating position in the vehicle seat. This enables optimized tuning and adjustment of the vehicle's restraint systems, ignition timing, ignition logic, etc., and allows for the output of information and error messages to occupant 1, particularly a driver, to correct their seating position and the positioning of the seat belt 2 on occupant 1's body. The tuning and output, especially of a warning message, constitute action A.

[0037] To classify an occupant 1 and their respective seating position, as well as to determine the path of the seat belt 2 on the occupant 1's body, a predictive model P trained using artificial intelligence is stored in a control unit 3, particularly for controlling the vehicle's restraint systems. The predictive model P is trained specifically using machine learning, whereby the predictive model P was and is being further trained using data from a conducted occupant study and / or image data of current vehicle interiors and / or data from an occupant simulation.Through the learning behavior of an algorithm for evaluating captured two-dimensional image data and captured three-dimensional image data, the control unit 3 is able to process a wide variety of input variables El to En in combination and initiate an action A with a significantly lower susceptibility to error in the vehicle.A seatbelt extension detection system for the seatbelt 2 assigned to occupant 1 offers, particularly in conjunction with captured image data and other input variables El to En, the possibility of more accurately representing the actual proportions of occupant 1, their seating position, the path of the fastened seatbelt 2 on their body, and their behavior during vehicle operation through comprehensive laboratory and / or subject studies and the use of artificial intelligence. This allows for the initiation of actions A that correct the seating position of occupant 1 and / or the path of the seatbelt 2 on their body. Particular emphasis is placed on optimizing the individual adjustment of the restraint systems to occupant 1.

[0038] The input variables El to En of the prediction model P for classifying occupant 1, in particular for determining his three-dimensional proportions, his seating positions and the course of the safety belt 2 on the body of occupant 1, are listed below.

[0039] An input variable El is a detected seatbelt buckle status, in particular when a seatbelt buckle tongue arranged on the seatbelt 2 is inserted into a seatbelt buckle assigned to this seatbelt 2.

[0040] An input variable E2 forms a belt extension length of the safety belt 2 determined on the basis of signals recorded from a sensor of a belt extension detection system, whereby the belt extension length is determined as an absolute value.

[0041] An input variable E3 of the prediction model P is an optical evaluation based on captured image data with regard to the occupant 1 and his seating position, wherein an input variable E4 is a vehicle state and / or an activation state of at least one assistance system of the vehicle.

[0042] Furthermore, the determined body weight of occupant 1 is fed into the prediction model P as input E5, whereby the body weight is determined, for example, using signals from a weight detection mat and / or another suitable measuring unit. Additionally, an input E6 of the prediction model P represents the course of the seat belt 2 determined from acquired image data, and a detection of the seating position also forms an input E7 of the prediction model P.

[0043] In particular, three-dimensional proportions of the occupant 1, the course of the safety belt 2 on the body of the occupant 1 and a three-dimensional sitting position of the occupant 1 are determined using the trained prediction model P and the input variables El to En supplied to it.

[0044] A tilt of the head 1.1 of occupant 1, for example, because they have fallen asleep in their vehicle seat, can only be detected using image data captured by a vehicle interior camera if the position of the head 1.1 noticeably changes in the direction of a transverse axis of the vehicle, depending on camera resolution tolerances. The path of the seat belt 2 on the body of occupant 1, and thus its extended length, where the path changes, for example, such that the seat belt 2 slips off a shoulder of occupant 1 or rests against the neck of occupant 1, cannot be detected.The changing length of the seat belt in terms of its speed, a change in length and possibly also a subsequent remaining in an immobile position provides information that the occupant 1 has significantly changed his seating position and that the seat belt 2 does not have an optimized path on the body of the occupant 1.

[0045] For example, occupant 1 might lift the seat belt 2 away from their chest for comfort while the vehicle is in transit. This lifting can be detected by the sensor for the belt extension detection system, which can identify a significant belt extension movement – ​​whether continuous, rapid, or temporary – and provide information about an unrealistic position of the seat belt 2.

[0046] A significant forward movement of occupant 1 in the direction of travel can be detected using three-dimensional image data captured by an interior vehicle camera. If the forward movement of the head 1.1 remains, but a reduction in the belt extension length is detected, it suggests that the seat belt 2 has slipped off one of occupant 2's shoulders. If the image data from an interior vehicle camera indicates that occupant 1 is not moving, but the seat belt 2 is clearly extended, the belt extension detection and temporal delta movements of the seat belt 2 reveal that the seat belt 2 is not in contact with occupant 1's body and therefore cannot fulfill its restraint function, or that occupant 1 has only briefly lifted the seat belt 2.

[0047] In particular, based on laboratory studies, different body and shoulder proportions, which influence the extension of the seat belt 2, can be optimally classified using different belt extension lengths while maintaining the same position of the occupant's head 1.1. If necessary, the classification of occupant 1 can be optimized compared to classification based solely on three-dimensional image data, using the optically captured seating position of occupant 1 as input E7 and the determined body weight of occupant 1 as input E5 of the predictive model P.

[0048] As shown in Figure 1, the lateral position of the head 1.1 of occupant 1, determined using three-dimensional image data, and an actual possible seating position of occupant 1 are depicted. Based on the acquired three-dimensional image data, an occupant's size can be determined in each axis direction of the vehicle coordinate system.

[0049] However, it is not possible to determine the shoulder position of occupant 1 based on the captured image data.

[0050] In a theoretical seating position S1, the occupant 1 has a seating position that is correct with regard to the measurement concept.

[0051] In a first practical seating position S2, the occupant 1 is not correctly positioned with respect to the vehicle seat; the occupant 1 sits relatively flat. In contrast, in a second practical seating position S3, the occupant 1 sits very upright, and there is a gap between the occupant 1's back and the backrest of the vehicle seat. In a further illustration in Figure 1, reference points B of seating positions S2 and S3 are shown in relation to seat reference points S of the theoretical seating position S1.

[0052] Even in the direction of travel, a shoulder position cannot be verified using three-dimensional image data captured by a vehicle interior camera positioned in front of the occupant 1.

[0053] A change in the position of the head 1.1 of the occupant 1 shown in Figure 2, in particular a change in the head angle with respect to a transverse axis of the vehicle, cannot be reliably detected using the three-dimensional image data captured, since only one point of the head 1.1 is used as the reference point B for the position of the head 1.1.

[0054] A lateral offset of the seating position in the direction of the vehicle's transverse axis can be detected using captured image data with an accuracy corresponding to a camera resolution, but does not provide any information about the actual seating position of occupant 1 on their vehicle seat.

[0055] The shoulder of occupant 1, as a two-dimensional contour, can potentially only be evaluated in the direction of the vehicle's transverse axis and in the direction of a vehicle's vertical axis, albeit with significantly higher computational effort. This method provides no information about a contact point BP between the shoulder and the seat belt 2 in relation to the occupant 1's seating position in the vehicle seat. Theoretically, the seat belt 2 runs optimally along the body of occupant 1, but it could slip off the shoulder or be too close to the occupant 1's neck. Since the head 1.1 of occupant 1 is largely in a theoretical position, an evaluation of the three-dimensional image data is insufficient in this regard.

[0056] Even a three-dimensional detection of the safety belt 2 in conjunction with the position of the head 1.1 is not unambiguous, since only the head 1.1 is considered for measurement.

[0057] Using two-dimensional image data to determine a predictable position of the head 1.1 of occupant 1 in three-dimensional space is comparatively inaccurate.

[0058] The seatbelt extension detection offers a way to relate laboratory data on body sizes, positions of the head 1.1 of occupant 1 determined using two-dimensional image data, a determined body weight of occupant 1 and the seatbelt extension length to the position of the shoulder of occupant 1, and to derive from this the percentile to which occupant 1 can be assigned, how and where occupant 1 sits approximately in relation to the vehicle seat.

[0059] Particularly in laboratory tests, belt extension lengths determined offer a way to optimally recognize information about the fastening status of the seat belt 2, its position relative to the shoulder of the occupant 1, its path, and the occupant's behavior during vehicle operation, all based on captured two-dimensional image data. Artificial intelligence can thus significantly improve the classification of the occupant 1 and their seating position, even using two-dimensional image data.

[0060] Reference symbol list

[0061] 1 occupant

[0062] 1.1 Head

[0063] 2 safety belt

[0064] 3 Control unit

[0065] A Action

[0066] B Reference point

[0067] BP contact point

[0068] El to En Input size

[0069] P Forecasting model

[0070] S Seat reference point

[0071] S1 to S4 seating position

Claims

Mercedes-Benz Group AG Patent claims 1. Method for classifying an occupant (1) on a vehicle seat in a vehicle and for detecting the seating position of the occupant (1) on the vehicle seat, wherein it is detected that the occupant (1) has fastened a seat belt (2) assigned to him / her and a body weight of the occupant (1) is determined, characterized in that - at least a belt extension length of a belt extension detection system and captured image data of at least a vehicle interior camera as input variables (El to En) are fed to a prediction model (P) trained using at least one artificial intelligence and - using at least one prediction model (P), three-dimensional proportions of the occupant (1) and / or a course of the seat belt (2) on the body of the occupant (1) and / or a three-dimensional seating position of the occupant (1) on the vehicle seat are determined or are determined. - a change in the course of the safety belt (2) on the body of the occupant (1) is determined based on a recorded speed of a change in the determined belt extension length.

2. Method according to claim 1, characterized in that the at least one prediction model (P) is trained using recorded data from a conducted occupant study and / or using image data of current vehicle interiors and / or using recorded data from an occupant simulation.

3. Method according to claim 1 or 2, characterized in that the belt extension length of the safety belt (2) is determined on the basis of detected signals from at least one sensor of the belt extension detection.

4. Method according to one of the preceding claims, characterized in that a forward displacement of the occupant (1) in the direction of travel of the vehicle is detected on the basis of captured three-dimensional image data and a slippage of the seat belt (2) from a shoulder of the occupant (1) caused by the forward displacement of the occupant (1) is determined on the basis of the decreasing belt extension length.

5. Method according to one of the preceding claims, characterized in that a vehicle state and / or an activation state of at least one assistance system is determined and supplied as input variables (El to En) to the prediction model (P).

6. Method according to one of the preceding claims, characterized in that a seat belt buckle status of the safety belt (2) assigned to the occupant (1) is detected and fed to the prediction model (P).

7. Method according to one of the preceding claims, characterized in that if it is determined by means of the prediction model (P) that the applied safety belt (2) generates an insufficient restraint effect on the occupant (1) in the event of need with regard to the occupant classification and / or its course on the body of the occupant (1), a warning message is issued in the vehicle.

Images