Method and setup for machine learning to determine the parking position of a motor vehicle in a parking space

Abstract

The invention relates to a method for determining a parking position (52, 54, 56, 58) of a motor vehicle (10) in a parking space (50), comprising the steps of: performing an image analysis using image data (42) of a passenger compartment (20) of the motor vehicle (10) by means of a machine learning device (30), wherein the machine learning device (30) detects during the image analysis where vehicle occupants (22) are arranged in the passenger compartment (20) and analyzes predetermined properties of the vehicle occupants (22) that influence how much space the vehicle occupants (22) need to exit the motor vehicle (10) and / or to enter the motor vehicle (10);Determining the parking position (52, 54, 56) using the machine learning device (30) taking into account the image analysis performed, so that the occupants (22) of the motor vehicle (10) each have an optimal amount of space available when entering and / or exiting the motor vehicle (10) after the parking process; wherein the machine learning device (30) was trained prior to the image analysis using machine learning based on at least one data set to perform such image analyses and, based thereon, to determine such parking positions (52, 54, 56, 58). Furthermore, the invention relates to a machine learning device (30) for carrying out such a method.

Description

[0001] The present invention relates to a method for determining the parking position of a motor vehicle in a parking space and to a machine learning device for carrying out such a method.

[0002] Automated parking of motor vehicles and the automated longitudinal and lateral guidance of the vehicle during the parking process by an assistance system are well-known. The assistance system steers the vehicle along a reference trajectory (lateral guidance) and regulates the speed (longitudinal guidance). Furthermore, multiple corrective maneuvers are possible to enable parking in tight spaces or to ensure the quality of the final position, such as the centering of the final position relative to the markings of the parking space or the orientation (forward / backward).

[0003] Various devices and methods for automated parking of motor vehicles are known from the prior art.

[0004] KR 10 2016 0 147 557 A concerns an automatic parking system that includes an interface unit which receives sensor information about the parking space. The system incorporates a camera to detect the number and position of the occupants and personal characteristics such as body weight, and is designed to execute the parking process in such a way as to facilitate convenient entry and exit for the occupants.

[0005] US 10,189,500 B2 relates to a parking assistance device comprising an occupant information processing unit, a peripheral information processing unit, and a parking adjustment unit. The occupant information processing unit includes an in-vehicle imaging unit and a weight detection unit. The parking assistance device considers features of the parking environment and personal characteristics such as the weight and height of all occupants as a function of the vehicle side in order to park the vehicle in a suitable position.

[0006] DE 10 2019 205 137 B4 concerns: The invention relates to a driver assistance system and a method for assisting a motor vehicle in parking a perpendicular parking space, which is laterally flanked by at least one parked neighboring vehicle. A camera system can be used to detect the neighboring vehicle and its position and orientation. Based on seatbelt buckle and / or seat occupancy detection information from the motor vehicle, the system also takes into account the number of people in the passenger compartment in order to prompt them to leave the vehicle before the parking maneuver is initiated. This is intended to prevent damage that can be caused, for example, by door collisions.

[0007] DE 10 2023 200 181 A1 relates to a system for facilitating entry and exit of a person from a vehicle. This system records the geometric characteristics of a vehicle occupant's body and, based on this information, determines a body length and / or circumference. Based on this determined length and / or circumference, a minimum door angle or a minimum width of an exit area in front of a door is calculated for a potential parking space. This information is then taken into account when automatically selecting a parking space and / or a predefined parking position.

[0008] DE 10 2020 207 442 A1 relates to a method for steering a vehicle into a parking space using an electronic parking aid. In this method, a parking space is detected, and, based on a sequence of steering movements, the vehicle is moved from its current position to a target position, the target position being dependent on at least one occupant parameter (e.g., weight).

[0009] DE 10 2020 126 322 A1 relates to a method in which a parking space is identified and a lateral distance is provided, if a vehicle is already parked, based on the door opening distance of the vehicle doors and / or the position of the occupants inside the vehicle. In addition, the parking position of the vehicle within the parking space is determined.

[0010] CN 1 14 347 999 A relates to a method for identifying a vehicle occupant based on several combined characteristics. Pressure values, seatbelt buckle signals, image data, and ultrasonic sensors are evaluated to determine the occupant's age, sex, and body type. The system distinguishes between infants, children, adult women, and medium-sized and tall men. The aim is to precisely classify vehicle occupants to improve safety and adapt restraint systems.

[0011] DE 10 2019 212 408 A1 relates to a method for determining the body weight and / or seating position of a vehicle occupant in a motor vehicle. In this method, a body characteristic of the vehicle occupant is determined by a feature module based on an image. Based on the determined body characteristic, a body weight and / or seating position is then determined by a conventional machine learning module.

[0012] The object of the invention is to provide a solution that enables particularly convenient entry and / or exit after parking in a parking space.

[0013] This problem is solved by the subject matter of the independent claims. Further possible embodiments of the invention are disclosed in the dependent claims, the description, and the figures. Features, advantages, and possible embodiments set forth in the description for one of the subject matter of the independent claims are to be regarded, at least analogously, as features, advantages, and possible embodiments of the respective subject matter of the other independent claims, as well as of any possible combination of the subject matter of the independent claims, optionally in conjunction with one or more of the dependent claims.

[0014] According to the invention, a method for determining the parking position of a motor vehicle in a parking space is provided. In a first step, an image analysis is performed using image data of the vehicle's passenger compartment by means of a machine learning device. During the image analysis, the machine learning device detects the position of vehicle occupants within the passenger compartment and analyzes predefined occupant characteristics that influence the amount of space required for entering and exiting the vehicle. In a second step, the machine learning device determines a parking position based on the image analysis, ensuring that the vehicle's occupants have an optimal amount of space for entering and / or exiting the vehicle after parking.Prior to image analysis using machine learning, the machine learning system was trained on at least one dataset to perform such image analyses and determine such parking positions based on them.

[0015] The invention has the advantage that, when determining the parking position, the vehicle occupants' seating positions and individual characteristics are taken into account; that is, all vehicle occupants are considered as human individuals. This takes into account that human individuals, due to their characteristics, have certain requirements regarding the parking position. This gives the occupants the feeling of a particularly human-centered parking process. The parking strategy also adapts to the seating occupancy; for example, if only the driver is in the vehicle, a position off-center is determined to allow for more comfortable entry and exit. Another advantage is that the correct orientation (forward / reverse) of the vehicle is selected. This allows neighboring vehicles to be considered as well, ensuring that their drivers can also enter and exit comfortably.A further advantage arises from the requirement that the presence of an interior camera in motor vehicles will become mandatory in many markets in the future. This means that hardware already installed in the vehicle can be used for the process, thus saving costs.

[0016] Non-limiting examples of motor vehicles within the scope of the invention are: passenger cars, trucks, or agricultural machinery. Non-limiting examples of such passenger cars include: small cars, compact cars, mid-size cars, luxury cars, sport utility vehicles (SUVs), station wagons, sedans, pickup trucks, light commercial vehicles, minibuses, minivans, and the like. A motor vehicle within the scope of the invention can, in particular, be a passenger car.

[0017] A parking space within the scope of the invention is a delimited area intended for parking a vehicle. The parking space can be defined by markings, adjacent vehicles or objects and can vary in shape, orientation and size, e.g. as a parallel, sloping or perpendicular gap to the roadway.Non-limiting examples of parking spaces within the scope of the invention include parallel parking spaces along the roadway, where the vehicle is parked parallel to the road; angled parking spaces, which are oriented at an angle to the roadway; right-angled parking spaces, which are arranged at a 90-degree angle to the roadway; longitudinal parking spaces in larger parking lots, where several vehicles park one behind the other; circular parking spaces, such as in multi-story car parks; parking spaces between two objects, for example, parked vehicles, walls, posts, and the like; diagonal parking spaces; parking spaces with limited space, which are characterized by a particularly narrow width or length; parking spaces in parking bays in curves, i.e., in areas where the roadway is curved; and irregularly shaped parking spaces, which do not have standard shapes due to their surroundings or boundaries, such as walls or uneven terrain.

[0018] In the context of the invention, a parking position can be understood as the orientation of a motor vehicle within a parking space relative to its boundaries. If the parking space is defined, for example, by right-angled markings, the vehicle can be aligned parallel to these markings. Alternatively, the motor vehicle can also be positioned at an angle within the parking space, i.e., not parallel to the markings. The motor vehicle can therefore also extend beyond the boundaries. In parking spaces that are not defined at right angles, the orientation of the motor vehicle can either correspond to the angle of the boundaries or deviate from it. A parking position in which the motor vehicle is positioned at an angle within the parking space is thus covered in the same way as a parking position in which the motor vehicle is aligned exactly in the center, i.e., centered with respect to all boundaries of the parking space.The boundaries of the parking space can be defined not only by markings, but also by adjacent motor vehicles or other objects that spatially define the parking space.

[0019] Image data within the scope of the invention can be understood as digital information representing visual content in the form of still images or moving images such as videos. Non-limiting examples of formats for such image data are JPEG, PNG, or BMP for still images and MP4, AVI, or MOV for moving image data. Image data can therefore include recordings produced by conventional still and / or video cameras, encompassing both static images and sequential recordings in the form of videos.

[0020] Machine learning within the scope of the invention can be understood as a subfield of artificial intelligence that deals with the development of algorithms and models that enable a system to learn from data and improve itself automatically without needing to be explicitly programmed. Non-limiting examples of learning methods include supervised, unsupervised, and reinforcement learning. Machine learning can be used, for example, in image analysis.

[0021] A machine learning system within the scope of the invention can be understood as a system that analyzes digital data based on algorithms and models to recognize patterns and relationships and derive predictions or decisions from them. Such systems typically include neural networks, decision trees, or support vector machines. Non-limiting examples of data formats processed include text, image data, or numerical data. The system can be trained using both supervised and unsupervised learning methods and can be specialized for different application areas, such as image analysis.

[0022] The phrase "predetermined characteristics of the vehicle occupants" within the context of the invention can be understood as features explicitly defined by an expert, i.e., a human worker. It can also be understood as features that are implicitly realized, i.e., not defined by an expert, and that are detected and recognized by the machine learning system. Thus, "predetermined characteristics" can also refer to implicitly defined or existing characteristics that influence the optimal amount of space within the context of the invention. For example, one of the vehicle occupants might be carrying an object such as a walking aid or a pet.It is also possible that a person's movements are perceived as comparatively cumbersome. Such characteristics can be implicitly interpreted as indications of a greater need for space when exiting and / or entering the vehicle, and thus belong to the category of "predetermined characteristics" by the vehicle occupants themselves.

[0023] The optimal amount of space within the scope of the invention can be understood in this context as the available space that allows a person to comfortably exit a vehicle without having to assume unnatural postures or contortions. This amount is not limited to a fixed length, as it depends on several factors, such as the required opening width of the vehicle door, the person's height and mobility, and the vehicle's geometry. The person's mobility may also depend on other factors, such as whether they are carrying items like a walking aid or a pet. Non-limiting examples of such influencing factors include the door size, the angle of the door opening, and the available space around the vehicle.

[0024] A displacement dimension within the scope of the invention can be understood as a dimension that describes an offset of the motor vehicle within the parking space. It can therefore be understood, for example, as a dimension of the deviation of the parking position of a motor vehicle with respect to a centered alignment of the motor vehicle in a parking space. The displacement dimension can, for example, have the dimension of a length, specified, for instance, in the unit millimeter [mm], centimeter [cm], decimeter [dm], or meter [m]. The displacement dimension can also be specified as a percentage deviation from a centered and / or parallel position, relative to the boundaries of the parking space. A prefix can be assigned to the respective displacement dimension to allow for its association with the side and the seating position in the motor vehicle.

[0025] Furthermore, according to the invention, the machine learning device comprises an artificial neural network. In this context, an artificial neural network can be understood as a machine learning model based on a simplified simulation of biological neural networks and consisting of several interconnected artificial neurons. Such neurons are typically organized in layers, with input data being passed through and processed by these layers to recognize patterns or make predictions. Non-limiting examples of such network types include feedforward networks, recurrent neural networks, and convolutional neural networks (CNNs). Artificial neural networks are typically used in areas such as image recognition, natural language processing, and autonomous decision-making.One advantage of using an artificial neural network as a machine learning tool is its particularly high performance in tasks such as image analysis, which allows for a very fast parking position determination process. This results in significantly reduced waiting times for parking position determination, creating a parking process that feels more human-like. A further advantage is that this can also reduce potential waiting times for other road users.

[0026] Furthermore, according to the invention, the machine learning system performs feature extraction regarding the characteristics of the vehicle occupants during image analysis based on the image data, wherein the characteristics include age, height, width, weight, and / or body size. This has the advantage that the characteristics of the vehicle occupants can be described more specifically, thereby achieving greater accuracy in determining the parking position. In one embodiment of the invention, the machine learning system can recognize during image analysis whether a vehicle occupant is a small child. In another embodiment of the invention, the machine learning system can further recognize during image analysis whether a child seat is present in the vehicle.

[0027] Furthermore, according to the invention, the extracted properties of the vehicle occupants are categorized into defined classes using the machine learning device. These classes qualitatively describe the characteristics of the vehicle occupants, and a quantitative displacement measure is assigned to each class based on its specific characteristics. This displacement measure describes the offset of the vehicle within the parking space. Classes within the scope of the invention can therefore be understood as categories used to classify the properties of the vehicle occupants according to specific features, with these features arranged in a graded sequence or scale. Typically, such classes represent qualitative or quantitative properties of an object or entity, such as age, weight, or size.The classification of vehicle occupants is typically based on predefined criteria derived from the characteristics being evaluated, designed to simplify the classification process. In one embodiment of the invention, the characteristic of vehicle occupants "age" can be categorized into the classes "very young, young, medium, old, very old." Similarly, the characteristic of vehicle occupants "height" can be categorized into the classes "very short, short, average, tall, very tall." Finally, the characteristic of vehicle occupants "width" can be categorized into the classes "very narrow, narrow, average, wide, very wide." For example, the machine learning system might recognize a vehicle occupant and determine that they are approximately 70 years old. This person would then be assigned to the "old" category.In one embodiment of the invention, a prefix may be assigned to the respective displacement dimension in order to associate the displacement dimension with the side and the seating position in the motor vehicle. In another embodiment of the invention, the displacement dimension may also be divided into classes, for example, into the classes "slightly to the left, to the left, strongly to the left" and "slightly to the right, to the right, strongly to the right".

[0028] In one embodiment of the invention, the respective displacement measures can be multiplied by a confidence measure, wherein the confidence measure describes the degree of certainty with which the respective characteristics of the vehicle occupant in question were recognized by the machine learning device during feature extraction. In this context, the confidence measure can be understood as a confidence interval that indicates the degree of certainty with which a vehicle occupant's characteristic was correctly recognized, i.e., with what degree of certainty, for example, it was recognized that a vehicle occupant is very young or has been classified as "very young." In one embodiment of the invention, the confidence measure can be determined using a statistical model.In one embodiment of the invention, it may be provided that the confidence measure is specified by training the machine learning device by a human expert.

[0029] In one embodiment of the invention, it may be provided that when determining the parking position, the driver is asked for confirmation if the confidence level falls below a predefined value of 95%, 90%, 85%, or 80%.

[0030] In one embodiment of the invention, the machine learning system can determine the parking position of the vehicle by taking into account all associated displacement measurements, ensuring that an optimal amount of space is available for entering and / or exiting the vehicle after parking. In this context, it can be provided that all displacement measurements determined for the left side of the vehicle are added together, and all displacement measurements determined for the right side of the vehicle are added together, and then the two sums are compared. The parking position can then be determined based on the larger sum.If the sum of all displacement measurements for the driver's side of the vehicle is greater than the sum of all displacement measurements for the passenger side, the parking position is determined based on the sum of all displacement measurements for the driver's side; thus, the parking position for the driver's side is optimized. The parking position can also be determined based on the arithmetic or geometric mean, or the difference, of the sums for the driver's and passenger's sides, weighting both sides. This results in a parking position that represents a compromise between the optimal parking positions for each side of the vehicle.It can also be provided that the individual displacement measurements determined for the right and left sides of the vehicle are not summed for each side, but rather that a single, maximum displacement measurement, determined for each seating position, is considered when determining the parking position. In other words, a displacement measurement can first be determined for each individual seating position, and then it can be checked whether there is a maximum displacement measurement for each side. Thus, the displacement measurements obtained for each seating position are not summed, but rather the highest displacement measurement determined for each side is checked. This can also include the case where both displacement measurements obtained for a vehicle side are the same, i.e.,For example, the displacement measurement obtained for the front passenger seat and the seat behind the front passenger seat can be the same. The two displacement measurements determined in this way for each side of the vehicle can then also be compared or related to each other, for example, based on the arithmetic or geometric mean or the difference between the two determined maximum displacement measurements for the driver's side and the front passenger side.

[0031] In one embodiment of the invention, the machine learning system can take into account at least one of the following criteria when determining the parking position: parking positions specified by a human operator; and / or parking positions determined in real-world studies with test subjects; and / or parking positions determined via questionnaires from test subjects; and / or parking positions determined from real-world fleet data; and / or parking positions that the driver of the vehicle typically prefers; and / or the individual exiting behavior of the driver of the vehicle. This offers the advantage of increased reliability in determining the parking position; that is, the machine learning system can further develop through machine learning in such a way that its reliability in determining a parking position is improved.

[0032] In one embodiment of the invention, it can be provided that vehicles parked in adjacent parking spaces and their parking direction are detected, and then, when determining the parking position, it is taken into account that the drivers of these vehicles have an optimal amount of space available for getting in and / or out of the vehicle after the parking process. This has the advantage of showing a certain degree of consideration for other road users, because they too then have more space available for getting in and out, while also reducing the risk of one of the vehicles being damaged, for example, by a door hitting another vehicle.

[0033] In one embodiment of the invention, the machine learning system can also take into account the arrangement of at least one vehicle adjacent to the parking space when determining the parking position. In this context, it can be provided that a parking position is determined in which the vehicle is parked at an angle in the parking space, for example, if a vehicle adjacent to the parking space is positioned at an angle. This has the advantage that the machine learning system also considers parking situations that deviate from an ideal.

[0034] In one embodiment of the invention, the machine learning system can check the arrangement of a child seat in the vehicle and take this information into account when determining the parking position to facilitate loading or unloading a child. In this context, it can be provided that the system detects the presence of a child seat in the vehicle even when no child is present, but, considering the vehicle's typical movement patterns, determines that a child is likely to be picked up and loaded into the vehicle after parking. This can then be taken into account when determining the parking position.This system can take into account that loading and unloading a vehicle with a child typically requires more space, for example, because the door needs to be fully opened and the adult needs a certain amount of room to maneuver. This has the advantage of determining a more predictive parking position even with an empty child seat, which offers a significant convenience benefit.

[0035] In one embodiment of the invention, the machine learning system can determine a parking position where, according to at least one predefined criterion, all vehicle occupants can easily exit the vehicle. Alternatively, one of the following actions can be performed: issuing a prompt for all vehicle occupants except the driver to exit before parking; or issuing a prompt for all vehicle occupants to exit, after which a parking maneuver is performed remotely, semi-automatically, or fully automatically. In this context, one criterion can include that the parking space is so narrow that, after the parking maneuver, it is no longer possible to open the doors without damaging the vehicle. This has the advantage of providing the driver and / or occupants with further options for determining a parking position in cases where no suitable parking position can be found.

[0036] The inventive device for machine learning is designed to carry out the inventive method or possible embodiments of the inventive method.

[0037] According to the invention, a motor vehicle can also be provided that can have such a machine learning device. The motor vehicle according to the invention further comprises a camera device which is designed to capture the passenger compartment, i.e., the camera device can produce images of the entire passenger compartment and generate image data from them.

[0038] Further features of the invention may become apparent from the following description of the figures and from the drawings. The features and combinations of features mentioned above in the description, as well as the features and combinations of features shown below in the description of the figures and / or in the figures themselves, can be used not only in the combinations specified, but also in other combinations or individually, without departing from the scope of the invention.

[0039] The drawing shows in: Fig. 1 a schematic representation of a motor vehicle which has a passenger compartment indicated by dashed lines, with a machine learning device and a camera device; Fig.2. A perspective view of the passenger compartment from the passenger door, showing the front part of the passenger compartment, the camera device, the steering wheel of the vehicle, and part of the driver. For illustrative purposes, a screen not belonging to the vehicle is also shown; Fig. 3 a schematic representation of the motor vehicle and a parking space as well as two motor vehicles arranged adjacent to the parking space; Fig. 4 a schematic representation of a motor vehicle, wherein the seating positions in the passenger compartment of the motor vehicle are arranged schematically.

[0040] Identical or functionally equivalent elements are marked with the same reference symbols in the figures.

[0041] A motor vehicle 10 is shown in a schematic representation in Fig.Figure 1 shows the motor vehicle 10 comprising a passenger compartment 20, a machine learning device 30, and a camera device 40. The camera device 40 provides image data 42 of the passenger compartment 20 to the machine learning device 30.

[0042] A perspective view of the passenger compartment 20 of the motor vehicle 10 from the viewpoint of the passenger door is shown in Fig.Figure 2 shows the front part of the passenger compartment 20, the camera device 40, the steering wheel of the motor vehicle 10, and part of the driver. The camera device 40 is configured to capture the passenger compartment 20; that is, the camera device 40 can produce images of the entire passenger compartment 20 and generate image data 42 from them, which are then transmitted to the machine learning device 30. For illustrative purposes, a screen is shown displaying an image of the passenger compartment 20 in which three vehicle occupants 22 are located; that is, the image data 42 contains an image of the three vehicle occupants 22 in the passenger compartment 20, and the image data 42 is transmitted to the machine learning device 30.

[0043] A parking space 50 and the motor vehicle 10 in front of it, with its rear facing parking space 50, are shown in a schematic representation in Fig.Figure 3 shows a parked vehicle 12 and 14 positioned to the right and left of parking space 50, respectively, each directly adjacent to parking space 50. Vehicles 12 and 14 are parked in a reverse-facing position, i.e., vehicle 12 is abutting parking space 50 with its passenger side, and vehicle 14 with its driver's side. Four possible parking positions 52, 54, 56, and 58 are also shown, their centers indicated by dashed lines.

[0044] A schematic representation of motor vehicle 10 is shown in the diagram. Fig. Figure 4 shows seating positions 60, 62, 64, 66 in the passenger compartment 20 of the motor vehicle 10 schematically indicated by boxes, where 60 is the seating position of the driver, 62 is the seating position of the front passenger, 64 is the seating position behind the seating position of the driver and 66 is the seating position behind the seating position of the front passenger.

[0045] In a first step of the procedure for determining the parking position 52, 54, 56, 58 of the motor vehicle 10 in the parking space 50, an image analysis is performed on the image data 42 of the passenger compartment 20 of the motor vehicle 10 using the machine learning device 30. During the image analysis, the machine learning device 30 detects where the vehicle occupants 22 are arranged in the passenger compartment 20. Predefined characteristics of the vehicle occupants 22 are analyzed, which influence how much space the vehicle occupants 22 need to exit and / or enter the motor vehicle 10. Prior to the image analysis, the machine learning device 30 was trained using machine learning based on at least one dataset to perform such image analyses; that is, the machine learning device 30 can recognize that there are three vehicle occupants 22 in the passenger compartment 20.Furthermore, the machine learning facility 30 can recognize that these vehicle occupants 22 are located in seat positions 60, 62 and 64.

[0046] The machine learning device 30 can then perform feature extraction regarding the characteristics of the vehicle occupants 22 based on the image data 42, where these characteristics can be age, height, and width. The extracted characteristics of the vehicle occupants 22 can then be classified by the machine learning device 30 into respective defined classes, which qualitatively describe the degree of the aforementioned characteristics of age, height, and width of the vehicle occupants 22, and depending on the degree of these characteristics, a quantitative displacement measure can then be assigned to each, which describes an offset of the vehicle 10 within the parking space 50.The characteristic "age" is divided into the classes "very young, young, medium, old, very old"; the characteristic "height" into the classes "very short, short, average, tall, very tall"; and the characteristic "width" into the classes "very narrow, narrow, average, wide, very wide".

[0047] This results in the features shown in the following table with their corresponding classes and displacement measures W1, W2 and W3: Feature 1 Feature 2 Feature 3 Old W1 height W2 Body width W3 very young 2 tiny 0 very narrow 0 young 1 small 0 narrow 0 medium 0 average 0 average 0 old 1 large 1 broad 1 very old 2 very large 2 very wide 2

[0048] The respective displacement measures can then be multiplied by a confidence measure, where the confidence measure describes the degree of certainty with which the respective properties of the vehicle occupant 22 in question were recognized by the machine learning device 30 during feature extraction. In this case, vehicle occupants 22 are located in seating positions 60, 64, and 66, for each of which Feature 1, Feature 2, and Feature 3 were extracted. Each of these features, i.e., Feature 1, Feature 2, and Feature 3, therefore has a corresponding displacement measure W1, W2, and W3, which are then summed to obtain a total displacement measure Wtotal = W1 + W2 + W3 for each seating position.

[0049] The following tables list these features again separately for each seating position, with the respective confidence measure for each feature listed in the last row of each table for the present example: Seating position 60 (driver's seating position) Feature 1 Feature 2 Feature 3 Old W1 height W2 Body width W3 medium 0 average 0 narrow 0 80 % 70 % 85 %

[0050] From this, the displacement dimension for seat position 60 is calculated as follows: Wtot(position 60)=(W1×0.80)+(W2×0.70)+(W3×0.85)Wtot(position 60)=(0×0.80)+(0×0.70)+(0×0.85)=0 Seating position 64 (seating position behind driver) Feature 1 Feature 2 Feature 3 Old W1 height W2 Body width W3 young 1 large 1 average 0 70% 75% 95%

[0051] From this, the displacement dimension for seat position 64 is calculated according to: Wtot(position 64)=(W1×0.70)+(W2×0.75)+(W3×0.95)Wtot(position 64)=(1×0.70)+(1×0.75)+(0×0.95)=1.45 Seating position 66 (seating position behind passenger) Feature 1 Feature 2 Feature 3 Old W1 height W2 Body width W3 old 1 small 0 very wide 2 90% 80% 95%

[0052] From this, the displacement dimension for seat position 66 is calculated according to: Wtot(position 66)=(W1×0.90)+(W2×0.80)+(W3×0.95) Wtot(position 66)=(1×0.90)+(0×0.80)+(2×0.95)=2.80

[0053] The obtained displacement measurements can now be added together, resulting in a total displacement of 1.45 (Wtotal (position 60) + Wtotal (position 64)) for the driver's side and a total displacement of 2.80 (Wtotal (position 66)) for the passenger side. Based on these summed displacement measurements, the most favorable parking position 54 for the passenger side of vehicle 10 can be determined.

[0054] If, on the other hand, it is intended that the individual displacement measurements determined for the right and left sides of the vehicle are not summed for each side, but rather that a single, maximum displacement measurement specific to each seating position is considered when determining the parking position, then it can first be determined which of the displacement measurements determined for each side is the highest. In the present example, this means that the highest displacement measurement determined for the right side of the vehicle is 2.8 and that the highest displacement measurement determined for the left side is 1.45. Instead of calculating a maximum sum per side, the determined displacement measurements can then be compared with each other.The two maximum displacement measurements for the driver's side and the passenger's side can be compared, for example, using the arithmetic or geometric mean or the difference between them. In this case, calculating the difference yields a value of 2.8 - 1.45 = 1.35. Therefore, parking position 58, not 54, would be a more suitable parking position. Parking position 54 takes into account a certain amount of space required for both the passenger and driver's sides, so that determining parking positions 52, 54, 56, and 58 represents a compromise between the calculated space requirements of both sides of the vehicle 10.

[0055] In a further step of the procedure for determining one of the parking positions 52, 54, 56, 58 of the motor vehicle 10 in parking space 50, the specific parking position 54 is selected from parking positions 52, 54, 56, 58 using the machine learning device 30, taking into account the image analysis performed. This ensures that the occupants of the motor vehicle 22 have an optimal amount of space available when entering and / or exiting the motor vehicle 10 after the parking process. The displacement measurements determined above, summed for the driver's side and the passenger's side respectively, can be considered in this process. A parking position favorable for the passenger side in parking space 50 is therefore parking position 54. Parking position 58, on the other hand, represents a compromise between a parking position favorable exclusively for the driver's side and one favorable exclusively for the passenger's side.The machine learning facility 30 was trained prior to image analysis using machine learning based on at least one data set to perform such image analyses and, based on this, to determine such parking positions 52, 54, 56, 58.

[0056] When determining parking positions 52, 54, 56, 58, the machine learning device 30 can consider at least one of the following criteria: The specification of parking positions 52, 54, 56, 58 by a human worker: Based on the determined displacement of 2.80 for the passenger side and the in Fig.Based on the parking space 50 shown in Figure 3, with parking positions 52, 54, 56, and 58, in conjunction with the orientation of the vehicle 10, a human operator can specify that parking position 54 is the position that provides optimal space on the passenger side for entering and / or exiting the vehicle 10 after parking. Alternatively or additionally, parking position 54 can also be determined based on real-world studies with test subjects and / or by using questionnaires to identify specific parking positions and / or from real fleet data. The machine learning system 30 can then consider frequently selected parking positions for the aforementioned combination of features and incorporate this knowledge into the determination of future parking positions, i.e., it can use this knowledge as a guide.Alternatively or additionally, parking position 54 can also be determined taking into account parking positions that the driver of the motor vehicle 10 has typically preferred in the past and / or are based on the individual exiting behavior of the driver of the motor vehicle 10.

[0057] It can further be provided that the vehicles 12, 14 positioned in the adjacent parking spaces and their parking direction are detected, and then, when determining one of the parking positions 52, 54, 56, 58, it is taken into account that the drivers of these vehicles 12, 14 have an optimal amount of space available for entering and / or exiting the motor vehicle 10 after the parking process. In the present case, parking position 54 would also be a favorable parking position for this purpose.

[0058] It can also be provided that the machine learning device 30 checks for the presence of a child seat in the vehicle 10 and takes this information into account when determining a parking position 52, 54, 56, 58 to facilitate loading or unloading a child. For example, if it is determined that a child seat is installed in seat position 66 and that a child is in it, this can be taken into account when determining the parking position 52, 54, 56, 58. If it is determined that a child seat is installed in seat position 66, but that no child is in it, it can be taken into account that, based on the usual movement profiles of the vehicle 10, a child is likely to be picked up and loaded into the vehicle 10 after the parking process.

[0059] If the machine learning system 30 cannot determine a parking position where all vehicle occupants 22 can exit without difficulty according to at least one predefined criterion, one of the following actions can be performed: issuing a request for all vehicle occupants 22 except the driver to exit before parking; or issuing a request for all vehicle occupants 22 to exit, after which a parking maneuver is performed remotely. In this context, it may be provided that one criterion includes the requirement that the parking space 50 is so narrow that it is no longer possible to open the doors after the parking maneuver without damaging the vehicle 10. For example, it is conceivable that the summed displacement measures determined for the respective sides of the vehicle 10 assume the same value, e.g., 5. That is to say,For example, a displacement dimension of 2 was determined for seat positions 60 and 64, and a displacement dimension of 4 for seat position 66, resulting in a sum of 4 for both the driver's and passenger's sides. Furthermore, the vehicle 14, positioned adjacent to parking space 50, borders parking space 50 on its driver's side. In such a case, several criteria would conflict, as an equally favorable parking position would ideally need to be determined for both the driver's and passenger's sides, and, as a further condition, slightly more space would need to be left to vehicle 14. Therefore, a request could be made for all vehicle occupants 22, except the driver, to exit vehicle 10, so that, as a compromise, parking position 52, for example, could be determined.Alternatively, the system can request that all vehicle occupants, including the driver, exit the vehicle so that the parking process can be carried out remotely, semi-automatically or fully automatically, and parking position 54 can be determined as the most favorable parking position. Reference symbol list 10 motor vehicle 12 Parked vehicle 14 Parked vehicle 20 passenger compartment 22 vehicle occupants 30 Machine Learning Facility 40 camera device 42 image data 50 parking spaces 52 Parking position, center 54 Parking position, left 56 Parking position, right 58 Parking position, between left and center 60 Seating position driver 62 Passenger seating position 64 Seating position behind driver 66 Seating position behind passenger

Claims

[1] Method for determining a parking position (52, 54, 56, 58) of a motor vehicle (10) in a parking space (50), comprising the steps: - Performing an image analysis using image data (42) of a passenger compartment (20) of the motor vehicle (10) by means of a machine learning device (30), wherein the machine learning device (30) detects during the image analysis where vehicle occupants (22) are arranged in the passenger compartment (20) and analyzes predefined properties of the vehicle occupants (22) that influence how much space the vehicle occupants (22) need to exit the motor vehicle (10) and / or to enter the motor vehicle (10); - Determining the parking position (52, 54, 56, 58) using the machine learning device (30) taking into account the image analysis performed, so that the occupants (22) of the motor vehicle (10) each have an optimal amount of space available when entering and / or exiting the motor vehicle (10) after the parking process; - wherein the machine learning device (30) was trained prior to image analysis using machine learning based on at least one data set to perform such image analyses and, based on these, to determine such parking positions (52, 54, 56, 58), wherein the machine learning device (30) performs feature extraction regarding the characteristics of the vehicle occupants (22) based on the image data (42) during image analysis, wherein the characteristics include age, height, width, weight and / or body fat, wherein the extracted properties of the vehicle occupants (22) are classified into respective defined classes using the machine learning device (30), which qualitatively describe the characteristics of the vehicle occupants (22), and wherein, depending on the characteristics of these, a quantitative displacement measure is assigned which describes an offset of the motor vehicle (10) within the parking space (50), and wherein the respective displacement measures are multiplied by a confidence measure, the confidence measure describing with what certainty the respective properties of the vehicle occupant in question (22) were detected by the machine learning device (30) during feature extraction. [2] Method according to claim 1, wherein the machine learning device (30) determines the parking position (52, 54, 56, 58) of the motor vehicle (10) taking into account all associated displacement dimensions, such that an optimal amount of space is available for entering and / or exiting the motor vehicle (10) after the parking operation. [3] Method according to claim 1 or 2, wherein the machine learning device (30) takes into account at least one of the following criteria when determining the parking position (52, 54, 56, 58): - Assignment of parking positions (52, 54, 56, 58) by a human worker; and / or - parking positions determined in real-world studies with test subjects (52, 54, 56, 58); and / or - parking positions determined via questionnaires to test subjects (52, 54, 56, 58); and / or - Parking positions determined from real fleet data (52, 54, 56, 58); and / or - Parking positions (52, 54, 56, 58) which the driver of the motor vehicle (10) typically prefers and / or the individual exiting behavior of the driver of the motor vehicle (10). [4] Method according to one of the preceding claims, wherein the machine learning device (30) also takes into account the arrangement of at least one motor vehicle (12, 14) adjacent to the parking space (50) when determining the parking position (52, 54, 56, 58). [5] Method according to any of the preceding claims, wherein the machine learning device (30) checks an arrangement of a child seat in the motor vehicle (10) and takes this information into account when determining the parking position (52, 54, 56, 58) to facilitate the loading or unloading of a child. [6] Method according to one of the preceding claims, wherein if the machine learning device (30) cannot determine a parking position (52, 54, 56, 58) at which all vehicle occupants (22) can exit without difficulty according to at least one predetermined criterion, one of the following actions is performed: - Issuing a request that all vehicle occupants (22) except for the driver should exit the vehicle before parking; or - Issuing a request for all vehicle occupants (22) to exit, after which a parking operation will be carried out remotely, semi-automatically or fully automatically. [7] Machine learning device (30) configured to perform a method according to any of the preceding claims.

Images