Cabin service control method and device, vehicle, storage medium and program product

Abstract

This application relates to the field of cockpit control technology, and particularly to a cockpit service control method, device, vehicle, storage medium, and program product, including: acquiring visual images and voice data within the cockpit; converting the voice data into text data; inputting the visual images and text data into a pre-trained visual language model, the visual language model outputting a visual language feature vector for each occupant; inputting the visual language feature vectors of all occupants into a graph neural network model, the graph neural network model outputting a relationship graph within the cockpit, each edge of the relationship graph labeled with a relationship type and intimacy score; identifying the target relationship type corresponding to the highest intimacy score in the relationship graph, determining the target open service and open permissions within the cockpit based on the target relationship type and intimacy score; and controlling the target open service based on the open permissions. This solves the problems of static relationship labels and low matching degree between service recommendations and occupant relationships in related technologies.

Description

Technical Field

[0001] This application relates to the field of intelligent cockpit technology, and in particular to a cockpit service control method, device, vehicle, storage medium and program product. Background Technology

[0002] With the rapid development of the intelligent vehicle industry, intelligent cockpits are gradually upgrading towards personalized and scenario-based adaptive services. User needs are no longer limited to basic audio-visual and air conditioning control, but rather they expect the system to dynamically match exclusive service solutions based on the real social relationships of the occupants in the vehicle, so as to achieve automatic control of the cockpit.

[0003] The relevant technologies mainly rely on static data such as historical travel records to build relationship graphs. The relationship tags are static, and the corresponding service output is mainly determined based on a single modal interaction, resulting in a low matching degree between service recommendations and passenger relationships. Summary of the Invention

[0004] This application provides a cabin service control method, device, vehicle, storage medium, and program product to solve problems such as static relationship tags and low matching degree between service recommendations and passenger relationships in related technologies.

[0005] The first aspect of this application provides a cabin service control method, comprising the following steps: acquiring visual images and voice data within the cabin; converting the voice data into text data; inputting the visual images and text data into a pre-trained visual language large model, wherein the visual language large model outputs the visual language features of each occupant; inputting the visual language feature vectors of all occupants into a graph neural network model, wherein the graph neural network model outputs a relationship graph within the cabin, wherein the relationship graph includes multiple graph nodes and edges, and each edge is labeled with a relationship type and a closeness score; identifying the target relationship type and closeness score corresponding to the highest closeness score in the relationship graph; determining the target open service within the cabin based on the target relationship type; and determining the open access permission of the target open service based on the closeness score; and controlling the target open service based on the open access permission.

[0006] Optionally, in one embodiment of this application, the processing procedure of the graph neural network model includes: creating multiple graph nodes according to the number of occupants, and establishing edges between every two graph nodes to form an initial relationship graph; calculating the interaction feature vector of the corresponding edge based on the visual language feature vectors of the two graph nodes connected by each edge; inputting the interaction feature vector into a predictive multilayer perceptron, wherein the predictive multilayer perceptron outputs the relationship type and intimacy score of the corresponding edge; and filling the initial relationship graph based on the relationship type and intimacy score of each edge to generate the final relationship graph.

[0007] Optionally, in one embodiment of this application, before identifying the target relationship type corresponding to the highest intimacy score in the relationship graph, the method further includes: identifying the number of times each pair of passengers travel together; correcting the relationship type and intimacy score of the corresponding edge labels in the relationship graph based on the number of times they travel together; and updating the relationship graph using the corrected relationship type and intimacy score.

[0008] Optionally, in one embodiment of this application, the visual language large model includes a visual feature extraction layer, a language feature extraction layer, and a cross-modal feature alignment layer. The visual feature extraction layer is used to extract visual features from visual images; the language feature extraction layer is used to extract language features from text data; the cross-modal feature alignment layer is used to map visual features and language features to a unified semantic space, calculate the similarity between visual features and language features, adjust the fusion weights of visual features and language features based on the similarity, and perform weighted fusion of visual features and language features based on the fusion weights to generate a visual language feature vector for the corresponding occupant.

[0009] Optionally, in one embodiment of this application, before controlling the target open service based on open permissions, the method further includes: obtaining current environment data; correcting the service parameters of the target open service based on the current environment data; and controlling the target open service based on open permissions and the corrected service parameters.

[0010] Optionally, in one embodiment of this application, before determining the target open service in the cabin based on the target relationship type, the method further includes: obtaining a relationship correction instruction input by the occupant; correcting the target relationship type based on the relationship correction instruction; and determining the target open service in the cabin based on the corrected target relationship type.

[0011] A second aspect of this application provides a cabin service control device, comprising: an acquisition module for acquiring visual images and voice data within the cabin; a first input module for converting the voice data into text data, inputting the visual images and text data into a pre-trained visual language large model, wherein the visual language large model outputs a visual language feature vector for each occupant; a second input module for inputting the visual language feature vectors of all occupants into a graph neural network model, wherein the graph neural network model outputs a relationship graph within the cabin, wherein the relationship graph includes multiple graph nodes and edges, and each edge is labeled with a relationship type and a closeness score; a determination module for identifying the target relationship type corresponding to the highest closeness score in the relationship graph, determining the target open service within the cabin based on the target relationship type, and determining the open access permission of the target open service based on the closeness score; and a control module for controlling the target open service based on the open access permission.

[0012] Optionally, in one embodiment of this application, the processing procedure of the graph neural network model includes: creating multiple graph nodes according to the number of occupants, and establishing edges between every two graph nodes to form an initial relationship graph; calculating the interaction feature vector of the corresponding edge based on the visual language feature vectors of the two graph nodes connected by each edge; inputting the interaction feature vector into a predictive multilayer perceptron, wherein the predictive multilayer perceptron outputs the relationship type and intimacy score of the corresponding edge; and filling the initial relationship graph based on the relationship type and intimacy score of each edge to generate the final relationship graph.

[0013] Optionally, in one embodiment of this application, it further includes: a first correction module, used to identify the number of times each pair of passengers travels together before identifying the target relationship type corresponding to the highest intimacy score in the relationship graph; correct the relationship type and intimacy score of the corresponding edge label in the relationship graph based on the number of times they travel together; and update the relationship graph using the corrected relationship type and intimacy score.

[0014] Optionally, in one embodiment of this application, the visual language large model includes a visual feature extraction layer, a language feature extraction layer, and a cross-modal feature alignment layer. The visual feature extraction layer is used to extract visual features from visual images; the language feature extraction layer is used to extract language features from text data; the cross-modal feature alignment layer is used to map visual features and language features to a unified semantic space, calculate the similarity between visual features and language features, adjust the fusion weights of visual features and language features based on the similarity, and perform weighted fusion of visual features and language features based on the fusion weights to generate a visual language feature vector for the corresponding occupant.

[0015] Optionally, in one embodiment of this application, it further includes: a second correction module, configured to obtain current environment data before controlling the target open service based on open permissions; correct the service parameters of the target open service based on the current environment data; and control the target open service based on open permissions and the corrected service parameters.

[0016] Optionally, in one embodiment of this application, it further includes: a third correction module, configured to obtain a relationship correction instruction input by the occupant before determining the target open service in the cabin based on the target relationship type; correct the target relationship type based on the relationship correction instruction; and determine the target open service in the cabin based on the corrected target relationship type.

[0017] A third aspect of this application provides a vehicle, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the cabin service control method as described above.

[0018] A fourth aspect of this application provides a computer-readable storage medium having a computer program or instructions stored thereon, which are executed by a processor to perform the cabin service control method as described above.

[0019] A fifth aspect of this application provides a computer program product, including a computer program or instructions, which, when executed, implement the cabin service control method as described in the above embodiments.

[0020] Therefore, this application has at least the following beneficial effects: This application's embodiments utilize a large visual-language model to process visual and voice data within the cabin, obtaining visual-language feature vectors for each occupant. These feature vectors are then input into a graph neural network model, which outputs a relationship graph within the cabin. By fusing multimodal data loops for analysis, the accuracy of occupant relationship analysis within the cabin is improved. The model identifies the target relationship type corresponding to the highest intimacy score in the relationship graph. Based on the target relationship type, target open services within the cabin are determined; based on the intimacy score, the access permissions for these target open services are determined; and then, access permissions are controlled based on these access permissions. By matching occupant relationship types with corresponding open services and occupant intimacy scores with access permissions, personalized cabin service automatic adjustment is achieved. This solves the technical problems of static relationship tags and low matching degree between service recommendations and occupant relationships in related technologies.

[0021] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0022] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a cabin service control method provided according to an embodiment of this application; Figure 2 This is a flowchart illustrating the cabin service control method provided according to an embodiment of this application. Figure 3 This is a flowchart of the relationship graph output according to the embodiments of this application; Figure 4 This is an example diagram of a cabin service control device provided according to an embodiment of this application; Figure 5 This is a structural schematic diagram of a vehicle provided according to an embodiment of this application. Detailed Implementation

[0023] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0024] With the development of smart cockpit technology, users' demand for personalized and scenario-based services is increasing. In existing technologies, social relationship perception mostly relies on static data (such as ride frequency and seat preference), lacking dynamic analysis of real-time interactive behaviors (such as dialogue content and body language). At the same time, the application of VLM (Vision-Language Model) in smart cockpits mostly focuses on single-modal interaction (such as voice control and image recognition), without fully integrating multimodal data to achieve deep reasoning of social relationships.

[0025] Related technologies only construct relationship graphs based on historical ride records, lacking real-time interactive analysis. Relationship tags are static, making them unsuitable for temporary scenarios (such as friends sharing a ride) and unable to capture dynamic relationships in real-time scenarios. Furthermore, they focus on single-modal interactions (such as voice control), introducing VLM to process cabin visual and voice data, but failing to associate social relationship types to optimize service output, resulting in low matching between service recommendations and passenger relationships. Furthermore, the privacy protection mechanisms in related technologies are inadequate, sensitive biometric data (such as facial and voice data) are easily leaked, they do not comply with data protection regulations and other standards, and lack a closed-loop feedback mechanism. After users manually adjust the service, the model cannot optimize itself (e.g., the system continues to recommend certain types of music even after users have manually turned them off multiple times), which can easily lead to user fatigue in the long run.

[0026] Therefore, this application provides a cabin service control method to solve at least one of the above-mentioned technical problems.

[0027] Specifically, Figure 1 This is a flowchart illustrating a cabin service control method provided in an embodiment of this application.

[0028] like Figure 1 As shown, the cabin service control method includes the following steps: In step S101, visual images and voice data inside the cockpit are acquired.

[0029] Visual images can be captured by an in-vehicle camera; voice data can be captured by an in-vehicle microphone.

[0030] Specifically, the hardware for acquiring visual images inside the vehicle cabin in this application embodiment includes four high-definition cameras, arranged as follows: the driver's side A-pillar camera focuses on the driver's facial expressions (such as frowning or smiling) and hand movements (such as the force with which the driver grips the steering wheel); the passenger side front camera captures the passenger's body movements (such as turning towards the driver or hand gestures); the rear left and right headrest cameras capture rear passenger interactions (such as children playing or elderly people resting); and the roof panoramic camera captures the overall cabin layout (such as seat distribution and item placement). Preprocessing: After distortion correction and illumination compensation, the images are input into the feature extraction module to ensure image quality in both low-light and high-light environments.

[0031] The hardware for collecting voice data can include an 8-channel microphone array deployed in the center of the ceiling. Adaptive filtering algorithms are used to cancel out engine noise and wind noise, maintaining voice clarity of ≥90% even at a vehicle speed of 80 km / h; and distinguishing the voice sources of the driver, front passenger, and rear passengers.

[0032] In step S102, the speech data is converted into text data, and the visual image and text data are input into the pre-trained visual language big model. The visual language big model outputs the visual language feature vector of each occupant.

[0033] The VLM model is a pre-trained model that includes a three-layer structure of visual, linguistic, and cross-modal alignment; the visual-linguistic features are a unified semantic vector obtained by weighted fusion of visual and linguistic features according to similarity.

[0034] It is understood that the embodiments of this application can convert voice data into text data, input visual images and text data into a pre-trained visual language big model, and the visual language big model outputs the visual language feature vector of each occupant.

[0035] The embodiments of this application can convert speech into text using an end-to-end ASR (Automatic Speech Recognition) model.

[0036] It should be noted that the embodiments of this application can also collect physiological data to assist in the determination of relationship type. The hardware for collecting physiological data may include: an 8×8 grid layout of seat pressure sensors integrated into the driver's seat, passenger seat, and rear seats to identify the occupant's sitting posture (such as leaning forward or backward); and heart rate sensors deployed on the steering wheel grip and rear armrest, using PPG (Photoplethysmography) to collect heart rate (accuracy ±1 beat / minute) and heart rate variability for emotion-assisted judgment (e.g., heart rate variability > 50ms indicates relaxation). Effective physiological features such as heart rate synchronicity, emotional arousal, and heart rate stability are extracted from the sensor data. Physiological data belongs to auxiliary enhancement features.

[0037] In one embodiment of this application, the visual language big model includes a visual feature extraction layer, a language feature extraction layer, and a cross-modal feature alignment layer. The visual feature extraction layer is used to extract visual features from visual images; the language feature extraction layer is used to extract language features from text data; the cross-modal feature alignment layer is used to map visual features and language features to a unified semantic space, calculate the similarity between visual features and language features, adjust the fusion weights of visual features and language features based on the similarity, and perform weighted fusion of visual features and language features based on the fusion weights to generate a visual language feature vector for the corresponding occupant.

[0038] Among them, visual features may include holding hands, eye contact, and children's chairs; linguistic features may include titles and conversation topics.

[0039] It is understood that the visual language large model in this application embodiment may include a visual feature extraction layer, a language feature extraction layer, and a cross-modal feature alignment layer. The visual feature extraction layer is used to extract visual features from visual images; the language feature extraction layer is used to extract language features from text data; the cross-modal feature alignment layer is used to map visual features and language features to a unified semantic space, calculate the similarity between visual features and language features, adjust the fusion weights of visual features and language features based on the similarity, and perform weighted fusion of visual features and language features based on the fusion weights to generate the visual language feature vector of the corresponding occupant. By aligning visual and language across modalities, conflicts between visual and language information are avoided.

[0040] Specifically, the visual feature extraction layer in this application embodiment may include a ResNet50 (Residual Network 50 layers) model, the language feature extraction layer may include a BERT (Bidirectional Encoder Representations from Transformer) model, and the cross-modal feature alignment layer may include a CLIP (Contrastive Language-Image Pre-training) model. The processing procedure of the large visual language model in this application embodiment includes: Visual features were extracted using the ResNet50 model to extract 2048-dimensional image features, with a focus on body movements, such as holding hands (distance between key hand points < 10cm), hugging (overlapping area of ​​torso > 30%), and sitting posture (close relationships tend to involve sitting side by side); facial expressions, such as smiling (angle of mouth > 15°), frowning (distance between eyebrows < 5cm), and eye contact (duration of eye contact > 2s).

[0041] The BERT speech feature model extracts 768-dimensional semantic features from the speech-to-text results, focusing on semantic features such as titles (dear, Mr. / Ms. Wang, etc.), topic words (child, project, etc.), relationship keywords (family, friends, clients, etc.), and scene keywords (dining, meeting, travel, etc.).

[0042] Multimodal feature alignment is achieved using the CLIP model architecture: visual features are encoded into 512-dimensional vectors using an image Transformer, and linguistic features are encoded into 512-dimensional vectors using a text Transformer. Utilizing CLIP's contrastive learning mechanism, the distance between matching feature pairs (e.g., an image of "holding hands" and text about "couples") is minimized, while the distance between non-matching feature pairs is maximized, mapping visual and linguistic features to a unified semantic space (similarity threshold ≥ 0.85). When the similarity of multimodal features reaches the set threshold, alignment is considered effective, and the process proceeds to the next fusion stage. Example: the cosine similarity between the visual feature of "holding hands" and the linguistic feature of "couples" is 0.92, while the similarity with the linguistic feature of "colleagues" is only 0.31. During the feature fusion stage, the weights of different features are dynamically adjusted based on the current scenario: intimate scenarios prioritize visual and audio input, while business scenarios prioritize dialogue text. Simultaneously, key information related to relationships is automatically strengthened. Finally, all features are integrated into a single comprehensive vector for subsequent relationship classification.

[0043] The following example illustrates the process of generating visual language feature vectors.

[0044] Step 1: Extract the original 2048-dimensional visual features using ResNet50.

[0045] The camera captured footage showing the mother repeatedly looking back at the children in the back seat, with eye contact lasting 6 seconds, and a 22° smile on her lips. The child seat was also visible in the footage. The ResNet50 convolutional network outputs a 2048-dimensional original visual feature vector V0. The feature channel values ​​corresponding to "child seat, prolonged eye contact, and smile" in the vector are significantly increased, indicating that intimate visual signals of family life have been captured.

[0046] Step 2: Compress the image using a Transformer to a 512-dimensional visual vector.

[0047] The 2048-dimensional V0 is fed into CLIP's built-in image Transformer for dimensionality compression and semantic mapping, removing redundant pixel information, ultimately resulting in a unified dimension: visual vector V∈R. 512 (512 dimensions).

[0048] Step 3: Speech-to-text conversion + BERT extraction of 768-dimensional original language features.

[0049] The microphone captured the mother's voice: "Baby, let's get something to eat at the service area later." After ASR denoising and text transcription, the text is input into BERT to extract 768-dimensional semantic features T0 of the original text. The channel weights corresponding to the keywords "baby", "service area", and "family" in the vector are increased.

[0050] Step 4: Compress the text using a Transformer to a 512-dimensional language vector.

[0051] The 768-dimensional T0 is compressed using CLIP text Transformer encoding and aligned with the visual vector dimensions to obtain: language vector T∈R 512 (512 dimensions).

[0052] Step 5: CLIP similarity check to determine if the alignment is valid.

[0053] The cosine similarity between V and T is 0.88, which is greater than the threshold of 0.85, thus satisfying the fusion condition. If replaced with colleague-related text such as "projects, clients", the similarity is only 0.31, indicating that the alignment has failed and the text will not be merged.

[0054] Step 6: Dynamically weighted fusion of the scene to output the final 512-dimensional visual language feature vector.

[0055] In the current family intimacy scenario, the algorithm increases the weight of the visual vector V by 0.6 and the weight of the text vector T by 0.4. The weighted fusion formula is: Feature = 0.6 × V + 0.4 × T. After fusion and normalization, a 512-dimensional comprehensive visual language feature vector Feature is obtained.

[0056] In step S103, the visual language feature vectors of all occupants are input into the graph neural network model. The graph neural network model outputs a relationship graph within the cabin, which includes multiple graph nodes and edges, with each edge labeled with a relationship type and intimacy score.

[0057] Among them, GNN is a network model with passengers as nodes and pairs of passengers as edges, used to build relationships between two people; each passenger can be uniquely identified by facial features or seat position; the relationship graph is a topological structure graph composed of graph nodes and edges, and each edge is labeled with relationship type and intimacy score; relationship types can include lovers, family, friends, colleagues, strangers, etc., and intimacy score is 0-10, with 10 being the highest intimacy.

[0058] It is understood that, in the embodiments of this application, the visual language feature vectors of all occupants can be input into a graph neural network model, and the graph neural network model outputs a relational graph within the cockpit.

[0059] In one embodiment of this application, the processing procedure of the graph neural network model includes: creating multiple graph nodes according to the number of occupants, and establishing edges between every two graph nodes to form an initial relationship graph; calculating the interaction feature vector of the corresponding edge based on the visual language feature vectors of the two graph nodes connected by each edge; inputting the interaction feature vector into a predictive multilayer perceptron, wherein the predictive multilayer perceptron outputs the relationship type and intimacy score of the corresponding edge; and filling the initial relationship graph based on the relationship type and intimacy score of each edge to generate the final relationship graph.

[0060] The initial relationship graph is a blank topology graph that only builds nodes and variable topology connections, without filling in relationship types or intimacy scores; the interaction feature vector is a unique vector obtained by fusing the complete feature vectors of the two occupants at both ends of an edge, carrying comprehensive information on two-way interaction between the two people. It is understood that the graph neural network model in this application embodiment can generate corresponding nodes according to the actual number of passengers, generate connecting edges between each pair of passengers, generate interaction feature vectors based on the feature vectors of the graph nodes at both ends of each edge, and input the interaction feature vectors into the predictive multilayer perceptron. The predictive multilayer perceptron outputs the relationship type and intimacy score of the corresponding edge, and then fills the initial relationship graph with the relationship type and intimacy score of each edge to obtain the final relationship graph.

[0061] For example, if there are three occupants in the cabin—a father, a mother, and a child—three nodes N1, N2, and N3 are generated. Three connecting edges E12, E13, and E23 are generated between each pair of nodes. Each node has a built-in 512-dimensional fused feature vector. Taking edge E13 (father-child) as an example: the features of N1 and N3 are concatenated to obtain a multi-dimensional vector, which is then compressed into a 256-dimensional interaction feature vector through a fully connected layer. This interaction feature vector is input into a predictive multilayer perceptron. This perceptron contains two hidden layers (128 dimensions in the first layer and 64 dimensions in the second layer) and has a dual-branch output. The classification branch uses a Softmax activation function to output the probabilities of five social relationships, and the label corresponding to the maximum probability is taken as the relationship type of the edge (couple / family / friend / colleague / stranger). The regression branch is scaled to the 0-10 range after Sigmoid activation mapping and outputs a floating-point intimacy score.

[0062] The graph neural network model outputs a relationship type of "family" with an intimacy score of 7.5. Similarly, E12 (couples, 7.2) and E23 (family, 7.3) are calculated.

[0063] Each node pre-stores a 512-dimensional visual language feature vector obtained through VLM cross-modal fusion; for a single edge E13, the 512-dimensional feature vectors corresponding to N1 and N3 are concatenated element by element to obtain a 1024-dimensional combined vector, which is then reduced in dimensionality by a 256-dimensional fully connected layer to output a 256-dimensional interactive feature vector.

[0064] The 256-dimensional interactive feature vector is fed into a predictive multilayer perceptron, which contains two hidden layers (128-dimensional and 64-dimensional). It is set with two output branches: the classification branch calculates the probability of five types of relationships through Softmax, with the highest probability label being family; the regression branch is scaled to the 0~10 range after being mapped by Sigmoid, and outputs a closeness score of 7.5.

[0065] Using the same logic, calculate the interaction features, relationship types, and intimacy levels corresponding to E12 and E23 sequentially: E12 is a couple with an intimacy level of 7.2; E23 is a family with an intimacy level of 7.3. Fill the relationship types and intimacy scores of all three sets of edges back into the corresponding edges of the initial graph to generate the final relationship graph carrying complete reasoning results.

[0066] In this embodiment, the relationship tag and intimacy level are adjusted every 30 seconds based on new interaction data (such as dialogue content and changes in body movements).

[0067] In step S104, the target relationship type corresponding to the highest intimacy score in the relationship graph is identified, the target open service and open permission in the cabin are determined based on the target relationship type, and the open permission of the target open service is determined based on the intimacy score.

[0068] The open services may include ambient lighting, multimedia, seat adjustment, navigation, etc.; the open permissions are the scope of available functions and usage restrictions corresponding to different relationship levels.

[0069] It is understood that the embodiments of this application can identify the target relationship type corresponding to the highest intimacy score in the relationship graph, determine the target open service in the cabin based on the target relationship type, determine the open permission of the target open service based on the intimacy score, match the corresponding open service with the occupant's relationship type, and match the open permission with the occupant's intimacy score, so as to realize the automatic adjustment of personalized cabin services.

[0070] In one embodiment of this application, before identifying the target relationship type corresponding to the highest intimacy score in the relationship graph, the method further includes: identifying the number of times each pair of passengers travel together; correcting the relationship type and intimacy score of the corresponding edge labels in the relationship graph based on the number of times they travel together; and updating the relationship graph using the corrected relationship type and intimacy score.

[0071] Among them, the number of times they traveled together is the number of times the two passengers have traveled together in the past, which is stored locally and serves as a constraint on the long-term relationship.

[0072] It is understood that the embodiments of this application identify the number of times each pair of passengers travel together, correct the relationship type and intimacy score of the corresponding edge labels in the relationship graph based on the number of times they travel together, and update the relationship graph using the corrected relationship type and intimacy score. By introducing long-term data constraints, the limitations of graph neural network models that rely solely on instantaneous in-vehicle images and dialogues in the cabin are compensated for, thereby improving the accuracy of relationship type and intimacy score judgments.

[0073] The number of shared trips in this application embodiment can be obtained based on the anonymized travel records and passenger association information stored locally in the vehicle. A valid trip event is defined as a vehicle driving process that meets the following conditions: (1) The vehicle starts and stops, and the single driving distance is ≥5km; the driving time is ≥10 minutes; the identity of the passengers in the vehicle can be identified by facial features or manual tags by the user. (2) During each trip, facial features are collected by the in-vehicle camera and compared with the authorized passenger feature database stored locally in the vehicle. If a match is found, it is marked as the same passenger. (3) After each valid trip, the vehicle system automatically records the IDs of all passengers present. Every two passengers present at the same time is recorded as a "shared trip". (4) All travel records and passenger association information are stored locally in the vehicle and are not uploaded to the cloud. Only the anonymized passenger IDs and trip counts are stored, and the original facial images or identity information are not stored. Users can view, delete or reset the travel count statistics in the vehicle system at any time. (5) For passengers who have not enabled facial recognition, the relationship can be determined by combining the passenger tags manually added by the user with the number of trips, thus avoiding functional limitations.

[0074] It should be noted that the embodiments of this application can also combine other multimodal features other than the number of times they travel together to modify the relationship type and intimacy score. The specific relationship classification rules are shown in Table 1.

[0075]

[0076] In one embodiment of this application, before determining the target open service in the cabin based on the target relationship type, the method further includes: obtaining a relationship correction instruction input by the occupant; correcting the target relationship type based on the relationship correction instruction; and determining the target open service in the cabin based on the corrected target relationship type.

[0077] Among them, the relationship correction instruction is a manual correction instruction that is manually entered or selected by the occupant, used to correct erroneous occupant relationships.

[0078] It is understood that the embodiments of this application can modify the target relationship type based on the relationship modification instruction, and determine the target open service in the cockpit based on the modified target relationship type, so as to avoid service opening errors caused by incorrect relationship type judgment.

[0079] Specifically, this application embodiment can ask "Are you satisfied? (1-5 stars)" via voice or touch screen within 30 seconds after the service ends, for example: "Did the restaurant you just recommended meet your expectations? Please rate it." Relationship correction: Supports users to manually correct relationship tags, for example: "The system currently identifies it as 'friend,' do you want to correct it to 'family'?"; allows users to adjust service parameters in real time (such as brightening the lights or changing the song), and the system records the adjustment trajectory.

[0080] Furthermore, it should be noted that the embodiments of this application can ensure that data processing complies with data protection regulations and other standards through a triple mechanism of "data desensitization - access control - local computing".

[0081] Step 1: Facial images retain facial expression features (such as mouth corners and eyebrows) but blur identity information (removing iris and texture details), and use k-anonymization (the same feature corresponds to at least k=5 users); Voice data: sensitive information such as ID card number and home address is masked when converting speech to text; Location data storage precision is dynamically adjusted according to relationship type (couples: street level; colleagues: city level; strangers: province level).

[0082] Step 2: Set up a data access whitelist based on relationship type. Car owners can access all data; family members can access non-private data such as shared travel records; colleagues / strangers can only access public data; and third-party services can only access anonymized preference tags.

[0083] Step 3: Core inference (VLM feature fusion, relationship classification) is completed on automotive-grade chips, and the original data is not uploaded to the cloud; only the de-identified service optimization parameters (such as "couples' preferred restaurant types") are uploaded to the cloud for cross-vehicle collaborative optimization; the original data is automatically deleted after 7 days by default, and the de-identified feature data is retained for 30 days.

[0084] In step S105, the target open service is controlled based on open permissions.

[0085] This application embodiment processes visual and voice data within the cabin using a large visual-language model to obtain visual-language feature vectors for each occupant. These feature vectors are then input into a graph neural network model, which outputs a relationship graph within the cabin. By fusing multimodal data loops for analysis, the accuracy of occupant relationship analysis within the cabin is improved. The model identifies the target relationship type corresponding to the highest intimacy score in the relationship graph. Based on the target relationship type, it determines the target open services within the cabin, and based on the intimacy score, it determines the access permissions for these target open services. Furthermore, it controls the target open services based on the access permissions. By matching occupant relationship types with corresponding open services and occupant intimacy scores with access permissions, personalized cabin service automatic adjustment is achieved.

[0086] In one embodiment of this application, before controlling the target open service based on open permissions, the method further includes: obtaining current environment data; correcting the service parameters of the target open service based on the current environment data; and controlling the target open service based on open permissions and the corrected service parameters.

[0087] The current environmental data includes the temperature inside and outside the vehicle, light intensity, and weather; the service parameters include headlight brightness, air conditioning temperature, and navigation preferences.

[0088] It is understood that the embodiments of this application can modify the service parameters of the target open service based on the current environmental data, and control the target open service based on the open permissions and the modified service parameters, thereby avoiding the control of the cockpit with fixed parameters and improving the cockpit's adaptive adjustment capability.

[0089] The environmental data acquisition method in this application embodiment may include: vehicle-mounted GPS / BeiDou dual-mode positioning to obtain geographical location and surrounding point of interest types (such as restaurants, shopping malls, scenic spots); meteorological API interface to obtain weather conditions, precipitation, and temperature in real time; vehicle CAN bus to collect vehicle speed, driving mode, and remaining battery power; and system time module to record time periods (morning peak 7:00-9:00, lunch break 12:00-14:00, night 20:00-22:00) and date types (weekday / weekend / holiday), etc.

[0090] Specifically, this application embodiment can generate a combined solution by calling atomic service units based on relational reasoning results and environmental parameters. The core is to achieve precise matching of relation-environment-service. First, the permissions of callable service units (i.e., target available services and service permissions) are filtered based on relation type. Then, the unit configuration (i.e., service parameters) is adjusted in combination with environmental parameters. Finally, a complete scenario-based service solution is generated. The specific process includes: I. Relationship Filtering Service Unit and Permissions.

[0091] The atomic service units that can be invoked are filtered according to the passenger relationship type, and only the service units and permissions that are suitable for the current relationship are opened: High intimacy (couples / core family): all service units are opened; Medium intimacy (friends / colleagues): only the basic service units suitable for the scenario are opened; Low intimacy (strangers): only the necessary services such as basic navigation and air conditioning are retained.

[0092] II. Environmental adjustment service unit parameters.

[0093] Based on the selected service units and permissions, the system dynamically adjusts the unit configuration by combining real-time environmental parameters (time, weather, vehicle speed, navigation scenario, etc.).

[0094] Step 3: Conflict coordination and solution generation.

[0095] Service units are combined according to the rule of "security first > user command first > scenario first" to handle parameter conflicts and generate complete scenario-based service solutions. 1. Security-related services are not affected by relationship type and environment; 2. The driver's manual commands can directly override the system's recommended configuration; 3. In the event of multiple scenario conflicts, the current core relationship scenario takes precedence.

[0096] The in-vehicle cabin services are atomized and broken down into 20+ basic units, each containing configurable parameters, as shown in Table 2.

[0097]

[0098] The matching of different relationships, service units, and permissions is shown in Table 3.

[0099]

[0100] Furthermore, service parameters are dynamically adjusted in conjunction with environmental parameters and optimized through the PPO reinforcement learning model. Service composition: State space: relationship type, intimacy, weather, time, vehicle speed; Action space: the combination of service units (such as "navigation + music + lights"), as shown in Table 4.

[0101]

[0102] This application's solution overcomes the limitations of traditional static relationship judgment. It achieves real-time updates of relationship tags through multimodal fusion, improving recognition accuracy in temporary scenarios (such as friends sharing a ride). It employs the CLIP model to bridge the "visual-text" semantic gap and combines relationship type and environmental parameters to generate combined services, enhancing user satisfaction and reducing service response latency. Through local computation, data anonymization, and access control, the risk of personal information leakage is reduced, complying with data protection regulations and other standards.

[0103] In summary, the entire processing flow of the cockpit service control method in this application embodiment is as follows: Figure 2 As shown, firstly, multi-dimensional data within the vehicle is synchronously collected via a distributed sensor array, providing rich input for relationship reasoning. Secondly, VLM relationship reasoning employs a visual language model to fuse multimodal features, constructing a dynamic social relationship graph and outputting relationship types and intimacy indices. The process of outputting the relationship graph is as follows: Figure 3 As shown in the diagram. Subsequently, based on the relational reasoning results and environmental parameters, atomic service units are invoked to generate combined solutions, adapting to different relations and scenarios. Furthermore, the model's reasoning is optimized based on user feedback. Finally, a triple mechanism of local computation, data anonymization, and access control is employed to ensure the security of sensitive data.

[0104] The cabin service control method of this application embodiment is described below through specific examples.

[0105] Example 1: A couple going out in the rain at dusk.

[0106] Multimodal data acquisition: The visual camera detected that the driver and front passenger were holding hands and both had smiles on their faces (the angle of the corners of their mouths was greater than 15°); the voice acquisition recorded the conversation "The rain is too heavy, let's find a place to sit" and "How is that Western restaurant you mentioned last time"; the environmental GPS showed that it was located in the city's business district, and the weather API returned the information "moderate rain" at 18:30.

[0107] VLM relationship reasoning: Visual features are extracted to include hand-holding and smile vectors, which are matched with intimate interaction tags; voice features are analyzed to include the last time the couple talked about and the Western restaurant, which are associated with historical travel records (3 restaurant dates); VLM cross-modal fusion outputs the couple relationship (intimacy score of 8), with environmental tags of rainy day + evening + business district.

[0108] Contextualized service generation: The navigation service recommends Western restaurants that both parties have saved (1.2km from the current location, including indoor parking), and the route planning avoids flooded sections of road; the ambient lighting switches to a pink-purple gradient (1Hz breathing mode), the driver and passenger seat massage is activated (gentle mode), the air conditioning is set to 24℃ and automatically dehumidifies; a jazz playlist that both parties have saved is played, and the volume is dynamically adjusted according to the vehicle speed (currently 60km / h → volume 65dB).

[0109] Interactive feedback optimization: During service execution, the passenger manually switches the music to piano music → the system records "preference for piano music > jazz music" and updates the personal music tag; upon arrival, the voice feedback is "very satisfied" (5 stars) → the service rules of "rainy day + couple → Western restaurant + pink and purple lights" are strengthened.

[0110] Privacy protection: Facial images are used only for expression analysis and are automatically deleted after 7 days of local storage; restaurant recommendations are based on publicly shared favorites by both parties and do not involve uploading private chat content.

[0111] Example 2: Family high-speed travel scenario.

[0112] Multimodal data acquisition: The rear camera detected a child seat (including a 3-year-old child), and the parent frequently looked back to pay attention (the duration of eye contact was >5 seconds); the voice assistant collected information such as whether the baby wanted to watch cartoons and whether to stop at the service area ahead; physiological detection showed the child's heart rate to be 90 beats / minute (normal range) and the parent's heart rate to be 85 beats / minute (slightly high, indicating attention); the GPS showed a highway section, 50km from the service area, the weather API returned sunny, 28℃, time 10:30, and vehicle speed 110km / h.

[0113] VLM Relationship Reasoning: Visual features of child seat + parent's attention and voice features of baby's name + service area request are extracted, and the data is linked to 8 historical weekend family travel records; cross-modal alignment shows a visual and voice feature similarity of 0.88, indicating a family relationship; the relationship classification GNN outputs the family relationship with an intimacy score of 7.5 (parent-child); dynamic updates: due to continuous interaction, the intimacy score remains stable (without decay). Contextualized Service Generation: Matching and calling family + long-distance highway rule bases; Service package: Navigation plans service area stops in advance, prioritizing smooth road sections; rear seat temperature is set to 25℃ (1℃ higher than the front seat), and seat ventilation is switched to low; the rear screen plays animation A, with simultaneous silent subtitles in the front; when 10km away from the service area, a voice prompt will announce that we are about to arrive at the service area and how long we need to stay; the PPO model prioritizes child comfort (weight 0.5), followed by driving safety (weight 0.4).

[0114] Interactive feedback optimization: Parents selected to stay on the touchscreen for 20 minutes and rated it 4 stars (the child really liked the animation); the child did not cry while watching the animation (positive feedback +2 points), and the parent did not adjust the temperature (positive feedback +3 points); the system recorded the 3-year-old child's preferred animation A, and will prioritize recommending similar content for the next family trip.

[0115] Privacy protection: Children's facial images are used only for age estimation and the raw data is not stored; service area recommendations are based on publicly available facility information (mother and baby rooms, children's play areas) and do not involve family privacy; all inference is performed locally and only the family's preferred service areas are uploaded.

[0116] According to the cabin service control method proposed in this application, visual and voice data in the cabin are processed by a large visual language model to obtain the visual language feature vector of each passenger. The visual language feature vectors of all passengers are then input into a graph neural network model. The graph neural network model outputs a relationship graph in the cabin. By fusing multimodal data loops for analysis, the accuracy of passenger relationship analysis in the cabin is improved, and the target relationship type corresponding to the highest intimacy score in the relationship graph is identified. The target open service in the cabin is determined based on the target relationship type, and the opening permission of the target open service is determined based on the intimacy score. Then, the target open service is controlled based on the opening permission. By matching the corresponding open service with the passenger's relationship type and the opening permission with the passenger's intimacy, the automatic adjustment of personalized cabin service is achieved.

[0117] Next, the cabin service control device proposed according to the embodiments of this application is described with reference to the accompanying drawings.

[0118] Figure 4 This is a block diagram of a cabin service control device according to an embodiment of this application.

[0119] like Figure 4 As shown, the cockpit service control device 10 includes: an acquisition module 100, a first input module 200, a second input module 300, a determination module 400, and a control module 500.

[0120] The system comprises the following modules: an acquisition module 100 for acquiring visual images and voice data within the cockpit; a first input module 200 for converting voice data into text data and inputting visual images and text data into a pre-trained visual language model, which outputs a visual language feature vector for each occupant; a second input module 300 for inputting the visual language feature vectors of all occupants into a graph neural network model, which outputs a relationship graph within the cockpit, wherein the relationship graph includes multiple graph nodes and edges, each edge labeled with a relationship type and intimacy score; a determination module 400 for identifying the target relationship type corresponding to the highest intimacy score in the relationship graph, determining the target open service within the cockpit based on the target relationship type, and determining the open access permission for the target open service based on the intimacy score; and a control module 500 for controlling the target open service based on the open access permission.

[0121] In one embodiment of this application, the processing procedure of the graph neural network model includes: creating multiple graph nodes according to the number of occupants, and establishing edges between every two graph nodes to form an initial relationship graph; calculating the interaction feature vector of the corresponding edge based on the visual language feature vectors of the two graph nodes connected by each edge; inputting the interaction feature vector into a predictive multilayer perceptron, wherein the predictive multilayer perceptron outputs the relationship type and intimacy score of the corresponding edge; and filling the initial relationship graph based on the relationship type and intimacy score of each edge to generate the final relationship graph.

[0122] In one embodiment of this application, the apparatus 10 of this application embodiment further includes: a first correction module.

[0123] The first correction module is used to identify the number of times each pair of passengers travel together before identifying the target relationship type corresponding to the highest intimacy score in the relationship graph; correct the relationship type and intimacy score of the corresponding edge label in the relationship graph based on the number of travels together, and update the relationship graph using the corrected relationship type and intimacy score.

[0124] In one embodiment of this application, the visual language big model includes a visual feature extraction layer, a language feature extraction layer, and a cross-modal feature alignment layer. The visual feature extraction layer is used to extract visual features from visual images; the language feature extraction layer is used to extract language features from text data; the cross-modal feature alignment layer is used to map visual features and language features to a unified semantic space, calculate the similarity between visual features and language features, adjust the fusion weights of visual features and language features based on the similarity, and perform weighted fusion of visual features and language features based on the fusion weights to generate a visual language feature vector for the corresponding occupant.

[0125] In one embodiment of this application, the apparatus 10 of this application embodiment further includes: a second correction module.

[0126] The second correction module is used to obtain current environment data before controlling the target open service based on open permissions; correct the service parameters of the target open service based on the current environment data; and control the target open service based on open permissions and the corrected service parameters.

[0127] In one embodiment of this application, the apparatus 10 of this application embodiment further includes: a third correction module.

[0128] The third correction module is used to obtain the relationship correction command input by the occupant before determining the target open service in the cabin based on the target relationship type; correct the target relationship type based on the relationship correction command; and determine the target open service in the cabin based on the corrected target relationship type.

[0129] It should be noted that the foregoing explanation of the cabin service control method embodiment also applies to the cabin service control device of this embodiment, and will not be repeated here.

[0130] According to the cabin service control device proposed in this application, the visual language feature vector of each passenger is obtained by processing visual and voice data in the cabin through a large visual language model. The visual language feature vectors of all passengers are then input into a graph neural network model. The graph neural network model outputs a relationship graph in the cabin. By fusing multimodal data loops for analysis, the accuracy of passenger relationship analysis in the cabin is improved, and the target relationship type corresponding to the highest intimacy score in the relationship graph is identified. The target open service in the cabin is determined based on the target relationship type, and the opening permission of the target open service is determined based on the intimacy score. Then, the target open service is controlled based on the opening permission. By matching the corresponding open service with the passenger's relationship type and the opening permission with the passenger's intimacy, the automatic adjustment of personalized cabin service is realized.

[0131] Figure 5 A schematic diagram of the structure of a vehicle provided in an embodiment of this application. The vehicle may include: The memory 501, the processor 502, and the computer program stored on the memory 501 and capable of running on the processor 502.

[0132] When processor 502 executes the program, it implements the cockpit service control method provided in the above embodiments.

[0133] Furthermore, the vehicle also includes: Communication interface 503 is used for communication between memory 501 and processor 502.

[0134] The memory 501 is used to store computer programs that can run on the processor 502.

[0135] Memory 501 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0136] If the memory 501, processor 502, and communication interface 503 are implemented independently, then the communication interface 503, memory 501, and processor 502 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0137] Optionally, in a specific implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface.

[0138] Processor 502 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0139] This application also provides a computer-readable storage medium storing a computer program or instructions thereon, which, when executed by a processor, implements the above-described cabin service control method.

[0140] This application also provides a computer program product, including a computer program or instructions, which, when executed, implement the above-described cabin service control method.

[0141] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0142] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0143] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0144] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0145] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A cabin service control method, characterized in that, Includes the following steps: Acquire visual images and voice data from inside the cockpit; The speech data is converted into text data, and the visual image and the text data are input into a pre-trained visual language large model. The visual language large model outputs the visual language feature vector for each occupant. The visual language feature vectors of all occupants are input into a graph neural network model, which outputs a relationship graph within the cabin. The relationship graph includes multiple graph nodes and edges, and each edge is labeled with a relationship type and intimacy score. Identify the target relationship type corresponding to the highest intimacy score in the relationship graph, determine the target open service in the cabin based on the target relationship type, and determine the open permission of the target open service based on the intimacy score; The target open service is controlled based on the aforementioned open permissions.

2. The cabin service control method according to claim 1, characterized in that, The processing steps of the graph neural network model include: Multiple graph nodes are created based on the number of occupants, and edges are established between every two graph nodes to form an initial relational graph. Calculate the interaction feature vector of the corresponding edge based on the visual language feature vectors of the two graph nodes connected by each edge; The interaction feature vector is input into a predictive multilayer perceptron, wherein the predictive multilayer perceptron outputs the relationship type and intimacy score of the corresponding edge; The initial relationship graph is populated based on the relationship type and intimacy score of each edge to generate the final relationship graph.

3. The cabin service control method according to claim 1, characterized in that, Before identifying the target relationship type corresponding to the highest intimacy score in the relationship graph, the process also includes: Identify the number of trips shared between any two occupants; Based on the number of shared trips, the relationship type and intimacy score of the corresponding edge labels in the relationship graph are corrected, and the relationship graph is updated using the corrected relationship type and intimacy score.

4. The cabin service control method according to claim 1, characterized in that, The large visual language model includes a visual feature extraction layer, a language feature extraction layer, and a cross-modal feature alignment layer, wherein... The visual feature extraction layer is used to extract visual features from the visual image; The language feature extraction layer is used to extract language features from the text data; The cross-modal feature alignment layer is used to map the visual features and the language features to a unified semantic space, calculate the similarity between the visual features and the language features, adjust the fusion weights of the visual features and the language features based on the similarity, and perform weighted fusion of the visual features and the language features based on the fusion weights to generate the visual-language feature vector of the corresponding occupant.

5. The cabin service control method according to claim 1, characterized in that, Before controlling the target open service based on the open permissions, the following is also included: Obtain current environmental data; Based on the current environment data, the service parameters of the target open service are corrected, and the target open service is controlled based on the open permissions and the corrected service parameters.

6. The cabin service control method according to claim 1, characterized in that, Before determining the target open service within the cabin based on the target relationship type, the process also includes: Obtain the relationship correction instruction input by the occupant; The target relationship type is modified based on the relationship modification instruction, and the target open service in the cockpit is determined based on the modified target relationship type.

7. A cabin service control device, characterized in that, include: The acquisition module is used to acquire visual images and voice data within the cockpit; The first input module is used to convert the speech data into text data, input the visual image and the text data into a pre-trained visual language large model, and the visual language large model outputs the visual language feature vector of each occupant. The second input module is used to input the visual language feature vectors of all occupants into the graph neural network model. The graph neural network model outputs the relationship graph in the cabin, wherein the relationship graph includes multiple graph nodes and edges, and each edge is labeled with the relationship type and intimacy score. The determination module is used to identify the target relationship type corresponding to the highest intimacy score in the relationship graph, determine the target open service in the cabin based on the target relationship type, and determine the open permission of the target open service based on the intimacy score; The control module is used to control the target open service based on the open permissions.

8. A vehicle, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the cabin service control method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, The computer program or instructions are executed by a processor to implement the cabin service control method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed, they implement the cabin service control method as described in any one of claims 1-6.

Images