Scene recognition methods, devices, computer equipment, storage media and products

Abstract

This application relates to a scene recognition method, apparatus, computer device, storage medium, and product. The method includes: acquiring multiple types of raw data and determining basic indicators based on the raw data; the raw data is data collected from the target vehicle; extracting semantic features of each basic indicator and generating semantic indicators based on the semantic features; aggregating the semantic indicators to determine combinations of semantic indicators, and generating scene indicators based on the combinations of semantic indicators; extracting features based at least on the scene indicators to obtain scene features, and determining a scene recognition result for the target vehicle based on the scene features. This method can improve the accuracy of scene recognition.

Description

Technical Field

[0001] This application relates to the field of automotive technology, and in particular to a scene recognition method, apparatus, computer equipment, storage medium, and product. Background Technology

[0002] With the development of intelligent cockpit technology, vehicles are gradually transforming from "functional centers" to "experience centers." Traditional cockpit control systems rely solely on raw sensor data (such as temperature, vehicle speed, and lighting status) for static control; however, with the development of artificial intelligence technology, large-scale models are gradually emerging to identify vehicle scenarios and achieve precise vehicle control based on these scenarios.

[0003] In related technologies, scene classification is usually based on preset types of driving scenarios. However, in actual driving, driving scenarios may be complex multimodal scenarios, which means that the preset driving scenarios cannot cover the actual driving scenarios, resulting in low scene recognition accuracy and affecting the subsequent vehicle decision-making effect. Summary of the Invention

[0004] Therefore, it is necessary to provide a scene recognition method, device, computer equipment, storage medium, and product that can improve the accuracy of scene recognition in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a scene recognition method, including:

[0006] Multiple types of raw data are acquired, and basic indicators are determined based on the raw data; the raw data is data collected from the target vehicle.

[0007] Extract the semantic features of each of the basic indicators, and generate each semantic indicator based on the semantic features;

[0008] Based on the semantic indicators, the indicators are aggregated to determine the combinations of semantic indicators, and based on the combinations of semantic indicators, the indicators for each scenario are generated.

[0009] Feature extraction is performed based on at least the scene indicators to obtain scene features, and scene recognition results for the target vehicle are determined based on the scene features.

[0010] In one embodiment, semantic features of each of the basic indicators are extracted, and each semantic indicator is generated based on the semantic features, including:

[0011] The basic indicators are encoded to obtain their semantic features;

[0012] Based on the semantic features of each of the basic indicators, feature distance is calculated to determine multiple combinations of basic indicators; each combination of basic indicators includes multiple basic indicators whose feature distance is less than a distance threshold.

[0013] Based on the basic indicators and their semantic features in the same basic indicator combination, the semantic indicators corresponding to the basic indicator combination are determined.

[0014] In one embodiment, based on each basic indicator in the same basic indicator combination and the semantic features of each basic indicator, the semantic indicator corresponding to the basic indicator combination is determined, including:

[0015] Based on the semantic features of each of the basic indicators in the same basic indicator combination, the indicator weight of each of the basic indicators is determined.

[0016] Based on the weights of each of the basic indicators, the basic indicators included in the combination of basic indicators are weighted and summed to obtain the semantic indicator corresponding to the combination of basic indicators.

[0017] In one embodiment, index aggregation is performed based on each of the semantic indicators to determine combinations of semantic indicators, and scene indicators are generated according to the combinations of semantic indicators, including:

[0018] Obtain an example of indicator aggregation; the example of indicator aggregation includes a reference scenario indicator and a combination of reference semantic indicators corresponding to the reference scenario indicator.

[0019] Based on the aforementioned indicator aggregation example, the indicator prediction model aggregates the semantic indicators to determine the combinations of semantic indicators; and generates scene indicators based on the combinations of semantic indicators.

[0020] In one embodiment, the method further includes:

[0021] Obtain a reference strategy example; the reference strategy example includes a reference scenario and a reference control strategy corresponding to the reference scenario;

[0022] The strategy generation model generates a strategy based on the reference strategy example and the scene recognition result to determine the target control strategy for the target vehicle.

[0023] Based on the target control strategy, target control commands are generated for the target vehicle.

[0024] In one embodiment, the method further includes:

[0025] Based on the scene recognition results and the target control strategy, an evaluation prompt message is generated;

[0026] Receive feedback information returned in response to the evaluation prompt information, and update at least one of the indicator prediction model and the strategy generation model based on the feedback information.

[0027] In one embodiment, multiple types of raw data are acquired, and various basic indicators are determined based on the multiple types of raw data, including:

[0028] Data is collected from the target vehicle based on the time synchronization signal to obtain various raw data.

[0029] The original data is processed to obtain processed data; the data processing includes at least one of filtering, normalization and standardization.

[0030] Based on the mapping relationship of basic indicator units, the processed data is mapped to the corresponding basic indicator units to obtain each basic indicator.

[0031] Secondly, this application also provides a scene recognition device, including:

[0032] The acquisition module is used to acquire various types of raw data and determine various basic indicators based on the raw data; the raw data is data collected from the target vehicle.

[0033] A semantic indicator construction module is used to extract the semantic features of each of the basic indicators and generate each semantic indicator based on the semantic features;

[0034] The scenario indicator construction module is used to aggregate indicators based on the semantic indicators, determine the combination of semantic indicators, and generate scenario indicators based on the combination of semantic indicators.

[0035] The scene recognition module is used to extract features based on at least the scene indicators to obtain scene features, and to determine the scene recognition result for the target vehicle based on the scene features.

[0036] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0037] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0038] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0039] The aforementioned scene recognition methods, devices, computer equipment, storage media, and products determine basic indicators through raw data, thereby representing the actual operating data of the target vehicle. Then, semantic indicators are determined through the semantic features of the basic indicators; that is, basic indicators with similar semantic features are fused to obtain semantic indicators. Semantic indicators establish semantic relationships between various raw data, realizing an abstract representation of semantic information from multiple raw data sets. Finally, indicators are aggregated based on semantic features to obtain scene indicators. The scene recognition result is determined based on the scene features of the scene indicators. This effectively integrates the semantic information of various semantic indicators to achieve accurate characterization of driving scenarios, making the obtained scene recognition results more consistent with actual driving scenarios, improving the accuracy of scene recognition, and providing a reliable basis for subsequent vehicle decisions, thereby enhancing the rationality and accuracy of vehicle control. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is an application environment diagram of the scene recognition method in one embodiment;

[0042] Figure 2 This is a flowchart illustrating a scene recognition method in one embodiment;

[0043] Figure 3 This is a flowchart illustrating step 201 in one embodiment;

[0044] Figure 4 This is a schematic diagram of the system architecture involved in one embodiment;

[0045] Figure 5 This is a structural block diagram of a scene recognition device in one embodiment;

[0046] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0048] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0049] The scene recognition method provided in this application embodiment can be applied to, for example, Figure 1 The application environment shown is illustrated. Terminal 101 communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be integrated onto server 102, or it can be located in the cloud or on another network server.

[0050] Terminal 101 can collect various raw data for the target vehicle, and then send the collected raw data to server 102. Server 102 obtains the raw data and determines various basic indicators based on the raw data. The raw data is data collected from the target vehicle. The semantic features of each basic indicator are extracted, and each semantic indicator is generated based on the semantic features. Based on each semantic indicator, the indicators are aggregated to determine the combination of each semantic indicator, and each scene indicator is generated based on the combination of each semantic indicator. At least based on the scene indicators, feature extraction is performed to obtain scene features, and the scene recognition result for the target vehicle is determined based on the scene features.

[0051] Terminal 101 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, and projection equipment. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. Server 102 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0052] In one exemplary embodiment, such as Figure 2 As shown, a scene recognition method is provided, which can be applied to... Figure 1 Taking the server in the example, the explanation includes the following steps 201 to 204. Wherein:

[0053] Step 201: Obtain various types of raw data and determine the basic indicators based on the various types of raw data.

[0054] The raw data refers to the data collected by the target vehicle, which is the data signal directly collected from the target vehicle's sensors or vehicle subsystems; for example, the vehicle speed and air conditioning temperature displayed on the vehicle's infotainment system, environmental images captured by the camera, and the on / off status data of the lights.

[0055] In some embodiments, basic indicators refer to indicators that can be directly mapped to the collected raw data. Further, determining basic indicators based on raw data can be achieved by pre-establishing mapping relationships between various basic indicator units and raw data. After collecting raw data, the raw data is mapped to the corresponding basic indicator units to obtain the basic indicators. For example, for the basic indicator unit "vehicle speed", the corresponding raw data is the vehicle speed collected by the speed sensor. After collecting the vehicle speed, the speed value of the vehicle speed is mapped to the basic indicator unit "vehicle speed" to obtain the basic indicator of vehicle speed.

[0056] Understandably, basic indicators should include at least an indicator name and indicator data; for example, for the basic indicator "vehicle speed", the indicator name can be "vehicle speed" and the indicator data can be the specific vehicle speed value "60".

[0057] It is understandable that, based on the basic indicators, including the indicator names, mapping various raw data to each basic indicator is essentially about giving the raw data specific semantic information through the basic indicators. This allows the raw data, which originally lacked a clear semantic orientation, to have a clear semantic expression, so as to facilitate the establishment of connections between data in the future.

[0058] In other embodiments, the basic metrics may also include attributes such as data unit, data sampling period, and data value range.

[0059] In some embodiments, the original data can be preliminarily verified by the preset attributes of the basic indicators; for example, the data value range included in the basic indicators can be used to preliminarily verify the original data; specifically, if the value of the original data is within the data value range included in the basic indicators, the data is determined to be normal and the corresponding basic indicators can be established; otherwise, the data is determined to be abnormal. In the case of determining that the data is abnormal, it is possible to choose to abandon the determination of basic indicators for the original data, thereby avoiding the impact of abnormal data on subsequent scene recognition.

[0060] Step 202: Extract the semantic features of each basic indicator and generate each semantic indicator based on the semantic features.

[0061] Semantic indicators refer to indicators obtained by fusing multiple basic indicators that have semantic relationships. For example, temperature, humidity, and noise are all basic indicators used to describe passenger experience. Therefore, multiple basic indicators corresponding to temperature, humidity, and noise can be fused to obtain semantic indicators about passenger experience.

[0062] In some embodiments, basic indicators can be encoded using embedding techniques to obtain the semantic features of the basic indicators.

[0063] In other embodiments, basic indicators can be encoded using a preset semantic template to obtain semantic features of the basic indicators; for example, the preset semantic template can be "indicator name, indicator value, value status judgment". Taking vehicle speed as an example, the semantic features obtained by the preset semantic template can be "vehicle speed, 20, low speed".

[0064] In some embodiments, semantic association analysis can be performed on the semantic features of each basic indicator through a semantic analysis model to determine the semantic features with semantic association. Then, the basic indicators corresponding to the semantic features with semantic association are combined to obtain semantic indicators.

[0065] Furthermore, to avoid the semantic relationships indicated by semantic indicators being too broad and losing their reference value, a maximum number of basic indicators corresponding to a semantic indicator can be set. When the semantic analysis model determines semantic features with semantic relationships, the number of semantic features with semantic relationships must not exceed the maximum number. By setting a maximum number, it is possible to effectively avoid extracting overly broad semantic relationships, making the semantic indicators have clear reference value. For example, the semantic features of all basic indicators related to the vehicle can be regarded as semantic features describing the vehicle state. If a semantic indicator of "vehicle state" is obtained based on all basic indicators, it is difficult to focus on key information because the semantic relationships it indicates are too broad, and its reference value for subsequent scene recognition is weak. If the maximum number is set to three, obviously, three semantic features with semantic relationships can determine more specific semantic relationships. For example, based on the window opening, seat angle, and driving mode, the semantic relationship can be determined as "all can affect the riding experience"; based on vehicle speed, noise, and wind speed, the semantic relationship can be determined as "all can affect the quietness inside the vehicle", etc.

[0066] In other embodiments, similar semantic features can be determined by calculating the feature distance between semantic features, and then the basic indices corresponding to the similar semantic features are combined to obtain semantic indices.

[0067] For example, for three basic indicators: temperature, humidity and noise, their semantic features are all related to vehicle comfort. Therefore, the indicator name of the semantic indicator can be determined as "comfort" based on temperature, humidity and noise. Furthermore, the indicator values ​​corresponding to temperature, humidity and noise can be summed to obtain the indicator data of the semantic indicator "comfort".

[0068] Step 203: Aggregate indicators based on each semantic indicator, determine the combination of each semantic indicator, and generate indicators for each scenario based on the combination of each semantic indicator.

[0069] Among them, scenario indicators refer to indicators obtained by aggregating multiple semantic indicators that have scenario-related relationships. For example, for nighttime driving scenarios, they are usually related to brightness, driving focus, and driving comfort. Therefore, multiple semantic indicators describing brightness, driving focus, and driving comfort can be aggregated to obtain the scenario indicator "nighttime driving comfort".

[0070] In some embodiments, cluster analysis can be used to discover the contextual relationships between multiple semantic indicators; furthermore, the rules for cluster analysis can be, for example, rules based on semantic similarity, rules based on business logic, etc.

[0071] Among them, the semantic similarity-based rule refers to the assumption that when multiple semantic indicators are highly similar in semantic connotation, there is a scene-related relationship between the multiple semantic indicators. For example, semantic analysis is performed on semantic indicators to obtain semantic analysis features. If the feature distance between the semantic analysis features of two semantic indicators is less than the distance threshold, the two semantic indicators are considered to be highly similar in semantic connotation, that is, there is a scene-related relationship between the two semantic indicators. In this case, these semantic indicators are aggregated to obtain a semantic indicator combination. For example, for the semantic indicators "comfort" and "quietness", both of which describe the passenger's feelings, it is possible to combine the two semantic indicators.

[0072] Business logic-based rules refer to aggregating semantic metrics based on pre-defined business requirement logic. Specifically, if the semantic content of two semantic metrics belongs to the same business requirement logic, the two semantic metrics are considered to have a scenario-related relationship. For example, in the state of "high-speed driving", the driver is required to maintain a high level of concentration. Therefore, the semantic metrics "high-speed driving" and "user status" can be aggregated.

[0073] In some embodiments, generating a scenario indicator based on a combination of semantic indicators can be achieved by predicting the indicator names of each semantic indicator in the semantic indicator combination to obtain the indicator name of the scenario indicator. For example, if the name of semantic indicator 1 is "comfort" and the name of semantic indicator 2 is "visibility", then predicting the indicator based on "comfort" and "visibility" can yield the indicator name of the scenario indicator "overall driving comfort". Alternatively, the indicator data of the scenario indicator can be obtained by linearly combining or weighted summing the indicator data of each semantic indicator. For example, if the indicator data of semantic indicator 1 "comfort" is 0.7 and the indicator data of semantic indicator 2 "visibility" is 200, then the "overall driving comfort" can be obtained as 2.7 (i.e., 2.7 = 0.7 + 200 / 100).

[0074] In some embodiments, the indicator prediction model can aggregate indicators based on various semantic indicators to determine combinations of semantic indicators, and generate various scene indicators based on the combinations of semantic indicators. The indicator prediction model is pre-trained. For example, it can be trained using multiple aggregation examples, which include multiple semantic indicators to be aggregated and the aggregated scene indicators. The indicator prediction model can learn the potential scene associations between semantic indicators and scene indicators through multiple aggregation examples. In this way, after obtaining the trained indicator prediction model, multiple semantic indicators are input into the trained indicator prediction model, and the trained indicator prediction model automatically mines the scene associations between semantic indicators, thereby outputting various scene indicators.

[0075] Step 204: Extract features based on scene indicators to obtain scene features, and determine the scene recognition result for the target vehicle based on the scene features.

[0076] Scene features refer to features that can reflect a specific driving scenario.

[0077] In some embodiments, feature extraction can be performed directly based on scene metrics to obtain scene features. For example, a neural network model can directly encode features based on scene metrics, mapping the scene metrics to a high-dimensional feature vector. Then, feature extraction can be performed based on the high-dimensional feature vector to obtain scene features. Alternatively, feature extraction can be performed based on both semantic and scene metrics. For example, a neural network model can encode features based on scene metrics and semantic metrics respectively, obtaining their corresponding feature vectors. Then, based on the correspondence between scene metrics and semantic metrics, the feature vectors corresponding to the scene metrics and semantic metrics are concatenated to obtain a concatenated vector. Finally, the concatenated vector is used for further processing. Scene features can be extracted from linear features; alternatively, they can be extracted based on basic indicators, semantic indicators, and scene indicators. For example, a neural network model can encode features based on basic indicators, semantic indicators, and scene indicators respectively to obtain their respective feature vectors. Then, based on the semantic association between basic indicators and semantic indicators, and the scene association between semantic indicators and scene indicators, the hierarchical relationship between the three indicators is determined. Subsequently, the feature vectors corresponding to the basic indicators, semantic indicators, and scene indicators are weighted and fused according to the hierarchical relationship to obtain a fused vector. Finally, feature extraction is performed based on the fused vector to obtain the scene features.

[0078] In some embodiments, the scene recognition model can extract features based on scene indicators to obtain scene features, and determine the scene recognition result for the target vehicle based on the scene features. The scene recognition model is pre-trained. For example, it can be trained using multiple scene recognition samples. Each scene recognition sample includes a sample scene label and at least one sample scene indicator corresponding to the sample scene label. The scene recognition model learns the potential association between the sample scene indicator and the sample scene label through the scene recognition samples. After training, by inputting the scene indicator into the trained scene recognition model, the scene recognition model can output the corresponding scene label.

[0079] Understandably, since scene metrics are determined based on semantic metrics, and semantic metrics are determined based on basic metrics, the final scene metrics fully consider the contribution of each basic metric to scene recognition in different dimensions, thus making the determined scene recognition results more consistent with the current actual driving scenario of the target vehicle. Taking the basic metric "vehicle speed" as an example, the method in this application can quantify its impact on multiple dimensions (such as "comfort", "quietness", "riding experience") in the process of determining the vehicle driving scenario, thereby improving the accuracy of scene recognition. In contrast, related technologies often directly determine scene prediction based on raw information such as "vehicle speed", ignoring its contribution to different dimensions of the driving scenario, resulting in poor scene recognition performance.

[0080] Furthermore, it is worth mentioning that there is a semantic relationship between basic indicators and semantic indicators, and a scenario relationship between semantic indicators and scenario indicators. Based on this, the relationship between basic indicators, semantic indicators, and scenario indicators can be determined. Further, after initially determining the relationship between basic indicators, semantic indicators, and scenario indicators, since the relationship is relatively stable, an indicator system can be constructed based on these indicators. If the type of the basic indicators remains unchanged, when the next scenario identification is required, only the indicator data of each indicator needs to be updated accordingly. If the basic indicators change, the indicator system can be reconstructed according to the above process to ensure the accuracy of the indicator system.

[0081] In the aforementioned scene recognition method, basic indicators are determined through raw data, which can then represent the actual operating data of the target vehicle. Next, semantic indicators are determined through the semantic features of the basic indicators; that is, basic indicators with similar semantic features are fused to obtain semantic indicators. Semantic indicators establish semantic relationships between various raw data, realizing an abstract representation of semantic information from multiple raw data sets. Finally, indicators are aggregated based on semantic features to obtain scene indicators. The scene recognition result is determined based on the scene features of the scene indicators. This method effectively integrates the semantic information of various semantic indicators to achieve accurate characterization of driving scenarios, making the obtained scene recognition results more consistent with actual driving scenarios, improving the accuracy of scene recognition, and providing a reliable basis for subsequent vehicle decisions, thereby enhancing the rationality and accuracy of vehicle control.

[0082] In one exemplary embodiment, such as Figure 3 As shown, step 201 includes steps 301 to 303, wherein:

[0083] Step 301: Data is collected from the target vehicle based on the time synchronization signal to obtain various raw data.

[0084] The time synchronization signal can be triggered based on the same clock source or through a synchronization trigger; the specific triggering method is not limited here.

[0085] It is understandable that data acquisition through time synchronization signals can ensure the consistency of sampling of various raw data in time, thereby facilitating subsequent combination and calculation based on various raw data.

[0086] Step 302: Perform data processing on the original data to obtain processed data; the data processing includes at least one of filtering, normalization and standardization.

[0087] Among them, filtering refers to removing interference and retaining useful signals from the original data containing noise or interference using a specific algorithm to improve data quality; normalization is to map data with different dimensions and value ranges to a unified interval, eliminate data differences, and facilitate data processing; standardization is to convert the original data into a standard data structure; for example, the standard data structure for processing data can be in JSON format to facilitate the subsequent construction of basic indicators.

[0088] Step 303: Based on the mapping relationship of basic indicator units, the processed data is mapped to the corresponding basic indicator units to obtain each basic indicator.

[0089] Among them, the basic indicator unit refers to the pre-set basic indicator type, such as "in-vehicle temperature", "vehicle speed", "ambient noise", "light intensity", etc.; further, the basic indicator unit mapping relationship refers to the relationship between the basic indicator type and the processed data, such as temperature sensor data corresponding to the basic indicator unit "in-vehicle temperature", speed sensor data corresponding to the basic indicator unit "vehicle speed", etc.

[0090] Mapping processed data to corresponding basic indicator units to obtain basic indicators means combining the effective data content of the processed data with the basic indicator units to obtain basic indicators; for example, if the speed sensor data is "60", it can be combined with the basic indicator unit "vehicle speed" to obtain the basic indicator "vehicle speed, 60".

[0091] In this embodiment, the processed data is mapped to the corresponding basic indicator units through the basic indicator unit mapping relationship to obtain each basic indicator. This can give the original data specific semantic information, so that the original data has a clear semantic expression, which is conducive to the subsequent mining of the relationship between data.

[0092] In an exemplary embodiment, step 202 may include the following steps one through three, wherein:

[0093] Step 1: Encode the basic indicators to obtain their semantic features.

[0094] The encoding can be either embedded encoding or semantic template encoding; the specific encoding method is not limited here.

[0095] In some embodiments, encoding basic indicators can be achieved by first expanding the text of the basic indicator based on preset rules to obtain the basic indicator description text, and then encoding the basic indicator description text to enrich the semantic features. For example, for the basic indicator "vehicle speed, 70", the corresponding basic indicator description text "current vehicle speed is 70, vehicle speed is high" can be generated, and then encoded according to the basic indicator description text.

[0096] Step 2: Calculate the feature distance based on the semantic features of each basic indicator to determine multiple combinations of basic indicators; each combination of basic indicators includes multiple basic indicators whose feature distance is less than the distance threshold.

[0097] The feature distance calculation can employ one or more of the following: cosine similarity, Euclidean distance, and Manhattan distance.

[0098] It is understandable that the smaller the feature distance between two semantic features, the higher the similarity between the two semantic features. In other words, it means that the two basic indicators have common semantic features. Therefore, multiple basic indicators with feature distance less than the distance threshold can be combined to obtain a basic indicator combination. Then, semantic indicators can be abstracted from the common characteristics of multiple basic indicators in the basic indicator combination.

[0099] It is understandable that a basic indicator can appear in different combinations of basic indicators.

[0100] Step 3: Based on the basic indicators and their semantic features in the same basic indicator combination, determine the semantic indicators corresponding to the basic indicator combination.

[0101] Specifically, by analyzing the semantic features of each basic indicator, we can abstract the semantic features of the semantic indicator, and then determine the indicator name based on these semantic features. For example, we can use unsupervised clustering or semantic embedding to aggregate basic indicators with similar semantic patterns into the same semantic cluster, with each cluster corresponding to a potential semantic concept. Subsequently, we assign an interpretable name to each semantic cluster as the indicator name of the semantic indicator. Through the indicator data of each basic indicator, we can obtain the indicator data of the semantic indicator. Finally, we combine the semantic name and the indicator data of the semantic indicator to obtain the semantic indicator.

[0102] For example, for the basic indicator combination "temperature, 24; humidity, 18; noise, 28", unsupervised clustering can be used to determine that all three indicators share the same semantic expression "passenger experience" based on their semantic features "temperature, humidity, noise". Therefore, "temperature, humidity, noise" can be clustered under the semantic cluster "passenger experience". Then, based on pre-learned prior knowledge, such as the common use of "comfort" to describe "passenger experience", the indicator name "comfort" can be assigned to "passenger experience", thus obtaining the semantic indicator "comfort". Afterwards, based on the temperature data... The values ​​"24", "18" for humidity, and "28" for noise are processed to obtain a comfort value of 0.7 (i.e., 0.7 = (24 + 18 + 28) / 100). Alternatively, multiple basic indicators and their corresponding comfort scores can be preset. Then, based on the range of values ​​for the basic indicators, the corresponding comfort score is determined, resulting in the semantic indicator "comfort, 0.7". It should be noted that this calculation method is only an illustrative example and does not constitute a limitation of this solution.

[0103] In other embodiments, semantic indicators can be obtained by semantic indicator model predicting semantic features based on reference semantic indicators, the combination of basic indicators corresponding to the reference semantic indicators, each basic indicator and the semantic features of each basic indicator.

[0104] In the above embodiments, the semantic similarity between each basic indicator is determined by the semantic feature distance, and then the semantic indicator is obtained based on multiple basic indicators with high semantic similarity, thereby realizing the semantic association of multiple basic indicators with high semantic similarity.

[0105] In one embodiment, step three may include steps five and six, wherein:

[0106] Step 5: Determine the weight of each basic indicator based on the semantic features of each basic indicator in the same basic indicator combination.

[0107] Among them, the indicator weight of the basic indicator refers to the degree of contribution of the basic indicator in the combination of basic indicators in the process of determining the semantic indicator.

[0108] In some embodiments, the semantic features of a semantic indicator can be determined first based on the semantic features of each basic indicator in the same basic indicator combination. For example, if the basic indicator combination includes basic indicator 1, basic indicator 2, and basic indicator 3, then semantic feature a can be determined based on the semantic features corresponding to each of basic indicator 1, basic indicator 2, and basic indicator 3, and the corresponding semantic indicator A can be determined based on semantic feature a. If the basic indicator combination includes basic indicator 1 and basic indicator 2, then semantic feature b can be determined based on the semantic features corresponding to each of basic indicator 1 and basic indicator 2, and the corresponding semantic indicator B can be determined based on semantic feature b. Then, the feature distance between the semantic features of each basic indicator and the semantic features of the semantic indicator is calculated. The closer the feature distance, the higher the indicator weight of the basic indicator.

[0109] Step 6: Based on the weights of each basic indicator, perform a weighted summation of the basic indicators included in the basic indicator combination to obtain the semantic indicator corresponding to the basic indicator combination.

[0110] Specifically, weighted summation of the basic indicators included in the basic indicator combination refers to weighted summation of the indicator data corresponding to each basic indicator in the basic indicator combination to obtain the indicator data of the semantic indicator, and thus obtain the complete semantic indicator.

[0111] In the above embodiments, the weight of each basic indicator is determined by the degree of contribution of each basic indicator in the process of determining the semantic indicator. Then, the semantic indicator is obtained by weighted summation based on the indicator weights, which can make the obtained semantic indicator more accurate.

[0112] In one embodiment, step 203 includes steps seven and eight. Wherein:

[0113] Step 7: Obtain an example of metric aggregation; the example of metric aggregation includes reference scenario metrics and the corresponding combination of reference semantic metrics.

[0114] It should be noted that there can be multiple indicator aggregation examples. These examples can provide prior knowledge to the indicator prediction model, thereby guiding the model in making indicator predictions.

[0115] Step 8: The indicator prediction model aggregates the semantic indicators based on the indicator aggregation example to determine the combination of semantic indicators; and generates the indicators for each scenario based on the combination of semantic indicators.

[0116] The indicator prediction model is a pre-trained neural network model. For example, it can be trained using multiple indicator aggregation examples. These examples include combinations of semantic indicators that need to be aggregated and the resulting scene indicators. Through these multiple indicator aggregation examples, the indicator prediction model can learn the potential scene relationships between semantic indicators and scene indicators. After obtaining the trained indicator prediction model, multiple semantic indicators are input into it, and the model automatically mines the scene relationships between the semantic indicators, thus outputting each scene indicator.

[0117] Understandably, through indicator aggregation examples, the indicator prediction model can not only learn the semantic relationship between scenario indicators and semantic indicators, but also the data relationship between the indicator data of scenario indicators and the indicator data of semantic indicators, thereby outputting the combination of semantic indicators and the scenario indicators corresponding to the combination of semantic indicators.

[0118] In the above embodiments, the indicator aggregation example is used to assist the indicator prediction model in predicting scene indicators, so that the indicator prediction model can capture the correlation between scene indicators and semantic indicators, thereby improving the prediction accuracy of the indicator prediction model.

[0119] In some embodiments, the scene recognition method further includes steps nine to eleven. Wherein:

[0120] Step 9: Obtain a reference strategy example; the reference strategy example includes a reference scenario and the reference control strategy corresponding to the reference scenario.

[0121] For example, the reference scenario is "nighttime driving scenario", and the corresponding reference control strategy is "turn on the ambient lighting, lower the air conditioning temperature, and lower the brightness of the vehicle display", etc.

[0122] Step 10: The strategy generation model generates a strategy based on the reference strategy example and the scene recognition results to determine the target control strategy for the target vehicle.

[0123] The scene recognition results can include scene semantic tags, such as "nighttime rest scene", "high-pressure driving scene", "social call scene", etc.

[0124] It is understandable that since the scenario is determined based on scenario indicators, and the scenario indicators are obtained by aggregating basic indicators layer by layer, that is to say, the basic indicators associated with the current scenario can be determined in reverse. Therefore, in some embodiments, the basic indicators, semantic indicators and scenario indicators obtained in the above process can also be input into the policy generation model. The policy generation model can then determine the basic indicators associated with the current scenario in reverse based on the relationship between the basic indicators, semantic indicators and scenario indicators, as well as the scenario recognition results. Finally, it can generate the corresponding target control policy based on the reference policy example and the basic indicators associated with the current scenario.

[0125] In some embodiments, user instructions can also be obtained, and the policy generation model can generate a policy based on the user instructions, reference policy examples, and scene recognition results to determine the target control policy for the target vehicle, so that the generated target control policy meets the scene requirements and user requirements.

[0126] Step 11: Generate target control commands for the target vehicle based on the target control strategy.

[0127] Among them, the target control command can be sent to the relevant execution unit (such as the air conditioning, audio and seat control module) through the bus protocol to execute the target control strategy.

[0128] In the above embodiments, a target control strategy for the target vehicle is obtained by referring to the strategy example and the scene recognition results, which improves the accuracy of the target control strategy. Then, the target control strategy is used to generate target control commands, thereby achieving precise control of the target vehicle.

[0129] In some embodiments, the scene recognition method further includes steps twelve to thirteen. Wherein:

[0130] Step 12: Based on the scene recognition results and target control strategy, generate evaluation prompt information.

[0131] The evaluation prompts may include evaluation entry points for scene recognition results and evaluation entry points for target control strategies.

[0132] In some embodiments, evaluation prompts may be displayed to users through pop-ups, voice broadcasts, or other means.

[0133] Step 13: Receive feedback information returned in response to the evaluation prompts, and update at least one of the indicator prediction model and strategy generation model based on the feedback information.

[0134] It is understandable that if the feedback information is not accurate enough, it means that the aforementioned scene recognition process is not accurate enough, or that the determined scene indicators are not accurate enough. Therefore, at least one of the indicator prediction model and the strategy generation model can be updated based on the feedback information to achieve self-learning of the model.

[0135] The model can be updated through reinforcement learning. Specifically, the scene recognition results, target control strategies, and feedback scores corresponding to feedback information in each interaction process are formed into a set of empirical data. The model parameters are adjusted using the empirical data to update the model.

[0136] In the above embodiments, at least one of the indicator prediction model and the strategy generation model is updated using feedback information. The model is iteratively optimized through online learning, thereby continuously improving the scene recognition effect and the control strategy generation effect.

[0137] In some of these embodiments, please refer to Figure 4 , Figure 4 The diagram illustrates a system architecture according to an embodiment of this application, specifically including:

[0138] The data acquisition module is used to acquire multi-dimensional raw data in real time from various vehicle subsystems and sensors, i.e., multi-source data. The collected data includes, but is not limited to: environmental data: interior and exterior temperature, humidity, noise, light intensity, and air quality; user status data: voice volume, facial expression, fatigue level, and posture information; device status data: air conditioning temperature, fan speed, window opening, seat angle, and headlight brightness; and driving status data: vehicle speed, gear position, fuel consumption, battery level, navigation route, and driving mode. The data acquisition module uses a unified timestamp synchronization mechanism and performs filtering, normalization, and standardization processing on the raw signals. The processed data is stored in a standard data structure, such as JSON format.

[0139] A unified data interface is used to provide a unified coding and interface standard for each module, enabling data to be shared and communicated between different functional domains, and providing a consistent indicator language.

[0140] The indicator modeling module is used to convert standardized data into indicators, including basic indicators, semantic indicators, and scenario indicators. Specifically, for basic indicators, standardized raw data is directly mapped to basic indicators based on pre-defined indicator mapping relationships, such as "in-vehicle temperature," "vehicle speed," "ambient noise," and "light intensity." Each basic indicator can include attributes such as name, unit, sampling period, and value range. Since basic indicators directly correspond to the raw data, automatic registration and updating of basic indicators can be achieved through the signal acquisition module.

[0141] For mid-level indicators, also known as semantic indicators, they can be obtained by normalizing, extracting semantic features, and calculating features from basic indicators, such as "comfort index", "brightness level", and "quietness index". For example, the semantic indicator "comfort index" can be obtained from three basic indicators "temperature", "humidity" and "noise", and can be expressed as comfort index = f(temperature, humidity, noise), where f can represent a linear combination, weighted average or machine learning model output.

[0142] For high-level indicators, namely scenario indicators, rule modeling or indicator prediction models can be used to aggregate semantic indicators to reflect the characteristics of complex scenarios; for example, scenario indicators include "driving focus", "relaxation atmosphere index" and "fatigue risk level".

[0143] In some embodiments, the indicator modeling module can also manage underlying indicators through an indicator registry and manage the calculation logic and upstream and downstream dependencies of indicators at each layer through an indicator dependency graph.

[0144] The indicator semantic modeling module is used to map indicators to a semantic space, enabling the indicator system to be understood by large models or algorithms. This module includes an indicator semantic encoding unit and an indicator semantic knowledge graph construction unit. The indicator semantic encoding unit uses embedding encoding technology or semantic template encoding to transform the definition, attributes, weights, and current values ​​of each indicator into vector representations. The indicator semantic knowledge graph construction unit refers to representing the above indicator modeling process through a knowledge graph and introducing the corresponding business strategies for each indicator to form a knowledge graph of "indicator-semantics-business strategy". For example, for the basic indicator "air conditioner fan speed", its corresponding semantic indicator is "energy consumption index". When the air conditioner fan speed is high, it is determined that the energy consumption index increases, and the corresponding business strategy is "reduce the air conditioner fan speed". Therefore, a graph structure of "higher air conditioner fan speed" - "increased energy consumption index" - "reduce air conditioner fan speed" can be constructed, which helps subsequent models refer to the knowledge graph to determine control strategies.

[0145] The scene recognition module is used to extract scene features based on indicators and determine the scene recognition result based on the scene features. Specifically, the scene recognition process includes: feature extraction, feature recognition, and scene label output. Feature extraction is used to select key feature vectors from at least the scene indicators, that is, feature vectors that contribute significantly to scene recognition. Feature recognition identifies the scene type through clustering algorithms or classification models. Scene label output is used to generate semantic labels based on the scene type, such as "nighttime rest scene," "high-pressure driving scene," and "social call scene."

[0146] In some embodiments, scene recognition supports two modes: rule-driven and model-driven. Rule-driven is suitable for explicit thresholds (such as temperature > 30°C and light intensity > 800 lux, which is determined as a "hot scene"); model-driven is suitable for complex or fuzzy semantic recognition (such as judging the "focused driving" state through multi-dimensional feature learning).

[0147] The business mapping module maps scene recognition results to vehicle control strategies, creating a closed loop from metrics to business execution. The module comprises a strategy generation unit and a control interface unit. The strategy generation unit generates the optimal business strategy (target control strategy) based on scene labels and metric status (i.e., the specific metric values). For example, if the scene is "relaxation atmosphere," the business strategy would be "adjust headlight brightness to 40%, play soft music, and recline the seat by 5°." The control interface unit converts the vehicle control strategy into vehicle control commands, which are then transmitted to relevant execution units (such as air conditioning, audio, and seat control modules) via bus protocols such as CAN (Controller Area Network), LIN (Local Interconnect Network), and Ethernet. This enables multi-task scheduling and priority management, ensuring driving safety and user experience.

[0148] The indicator management and feedback module is used to continuously maintain and dynamically optimize the indicator system. Specifically, this module supports the addition, modification, and deletion of basic indicators; it can record business execution results and user feedback for optimizing various large models; and it can continuously improve the accuracy of indicator prediction and scenario recognition through online learning or cloud training.

[0149] Through the aforementioned system architecture, the underlying physical signals scattered across different subsystems are transformed into unified semantic indicator expressions, achieving a continuous mapping from the signal layer to the semantic layer. Furthermore, through unified basic indicator definitions, encoding, and interface standards, data between functional domains can be shared and interconnected, providing a consistent indicator language for the intelligent cockpit. This allows different business modules to collaborate based on the same indicator system, providing standardized input for subsequent large-scale model inference and improving indicator interoperability and reusability. Simultaneously, through indicator aggregation and semantic modeling, complex scene features can be automatically identified from multi-dimensional data, forming a high-level scene indicator layer. This layer can express the non-linear relationships between driving scenarios, user states, and environmental features, thereby achieving more accurate scene classification and semantic understanding. Scene recognition accuracy is improved compared to traditional rule-based methods, and it can identify ambiguous, complex, or dynamically changing scenes (such as "focused driving" and "relaxed entertainment"). It supports unified scene perception through multi-modal input fusion (voice, sensor, vision). Finally, based on the indicator system and scene recognition results, the optimal control strategy can be dynamically generated to achieve intelligent linkage between indicators, scenes, and strategies; the control strategy changes from static rules to dynamic adaptation, significantly improving the level of intelligence; moreover, new devices or sensors can be quickly connected, only basic indicators need to be added, without the need to reconstruct the algorithm system; forming an enterprise-level or industry-level cockpit intelligent indicator standard system.

[0150] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0151] Based on the same inventive concept, this application also provides a scene recognition device for implementing the scene recognition method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more scene recognition device embodiments provided below can be found in the limitations of the scene recognition method described above, and will not be repeated here.

[0152] In one exemplary embodiment, such as Figure 5 As shown, a schematic diagram of a scene recognition device is provided. The scene recognition device 500 includes:

[0153] The acquisition module 501 is used to acquire various raw data and determine various basic indicators based on the raw data; wherein, the raw data is the data collected from the target vehicle.

[0154] The semantic indicator construction module 502 is used to extract the semantic features of each basic indicator and generate each semantic indicator based on the semantic features.

[0155] The scenario indicator construction module 503 is used to aggregate indicators based on various semantic indicators, determine the combination of semantic indicators, and generate various scenario indicators based on the combination of semantic indicators.

[0156] The scene recognition module 504 is used to extract features based on at least scene indicators, obtain scene features, and determine the scene recognition result for the target vehicle based on the scene features.

[0157] In one embodiment, the semantic index construction module 502 is specifically used to encode the basic indexes to obtain the semantic features of the basic indexes; to calculate the feature distance based on the semantic features of each basic index to determine multiple combinations of basic indexes; each combination of basic indexes includes multiple basic indexes whose feature distance is less than a distance threshold; and to determine the semantic index corresponding to the combination of basic indexes based on each basic index and the semantic features of each basic index in the same combination of basic indexes.

[0158] In one embodiment, the semantic index construction module 502 is specifically used to: determine the index weight of each basic index based on the semantic features of each basic index in the same basic index combination; and perform a weighted summation of the basic indexes included in the basic index combination based on the index weight of each basic index to obtain the semantic index corresponding to the basic index combination.

[0159] In one embodiment, the scenario indicator construction module 503 is specifically used to obtain indicator aggregation examples; the indicator aggregation examples include reference scenario indicators and reference semantic indicator combinations corresponding to the reference scenario indicators; the indicator prediction model performs indicator aggregation on each semantic indicator based on the indicator aggregation examples to determine each semantic indicator combination; and generates each scenario indicator based on each semantic indicator combination.

[0160] In one embodiment, the scene recognition device 500 further includes an instruction generation module for obtaining a reference strategy example; the reference strategy example includes a reference scene and a reference control strategy corresponding to the reference scene; the strategy generation model generates a strategy based on the reference strategy example and the scene recognition result to determine a target control strategy for the target vehicle; and a target control instruction for the target vehicle is generated based on the target control strategy.

[0161] In one embodiment, the scene recognition device 500 further includes an optimization module, which generates evaluation prompt information based on the scene recognition result and the target control strategy; receives feedback information returned in response to the evaluation prompt information; and updates at least one of the indicator prediction model and the strategy generation model based on the feedback information.

[0162] In one embodiment, the acquisition module 501 is specifically used to collect data from the target vehicle based on the time synchronization signal to obtain various raw data; to process the raw data to obtain processed data; the data processing includes at least one of filtering, normalization and standardization; and to map the processed data to the corresponding basic indicator units based on the basic indicator unit mapping relationship to obtain each basic indicator.

[0163] Each module in the aforementioned scene recognition device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0164] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores scene recognition data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a scene recognition method.

[0165] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0166] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0167] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0168] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method described above.

[0169] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0170] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0171] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0172] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A scene recognition method, characterized in that, The method includes: Multiple types of raw data are acquired, and basic indicators are determined based on the raw data; the raw data is data collected from the target vehicle. The basic indicators are encoded to obtain their semantic features; Based on the semantic features of each of the basic indicators, feature distance is calculated to determine multiple combinations of basic indicators; each combination of basic indicators includes multiple basic indicators whose feature distance is less than a distance threshold. Based on the semantic features of each of the basic indicators in the same basic indicator combination, the indicator weight of each of the basic indicators is determined. Based on the weights of each of the basic indicators, the basic indicators included in the combination of basic indicators are weighted and summed to obtain the semantic indicator corresponding to the combination of basic indicators. Based on the semantic indicators, the indicators are aggregated to determine the combinations of semantic indicators, and based on the combinations of semantic indicators, the indicators for each scenario are generated. Feature extraction is performed based on at least the scene indicators to obtain scene features, and scene recognition results for the target vehicle are determined based on the scene features.

2. The method according to claim 1, characterized in that, Based on the aforementioned semantic indicators, indicator aggregation is performed to determine the combinations of semantic indicators, and based on the combinations of semantic indicators, various scenario indicators are generated, including: Obtain an example of indicator aggregation; the example of indicator aggregation includes a reference scenario indicator and a combination of reference semantic indicators corresponding to the reference scenario indicator. Based on the aforementioned indicator aggregation example, the indicator prediction model aggregates the semantic indicators to determine the combinations of semantic indicators; and generates scene indicators based on the combinations of semantic indicators.

3. The method according to claim 2, characterized in that, The method further includes: Obtain a reference strategy example; the reference strategy example includes a reference scenario and a reference control strategy corresponding to the reference scenario; The strategy generation model generates a strategy based on the reference strategy example and the scene recognition result to determine the target control strategy for the target vehicle. Based on the target control strategy, target control commands are generated for the target vehicle.

4. The method according to claim 3, characterized in that, The method further includes: Based on the scene recognition results and the target control strategy, an evaluation prompt message is generated; Receive feedback information returned in response to the evaluation prompt information, and update at least one of the indicator prediction model and the strategy generation model based on the feedback information.

5. The method according to any one of claims 1 to 4, characterized in that, Acquire various raw data sets and determine basic indicators based on these raw data sets, including: Data is collected from the target vehicle based on the time synchronization signal to obtain various raw data. The original data is processed to obtain processed data; the data processing includes at least one of filtering, normalization and standardization. Based on the mapping relationship of basic indicator units, the processed data is mapped to the corresponding basic indicator units to obtain each basic indicator.

6. A scene recognition device, characterized in that, The device includes: The acquisition module is used to acquire various types of raw data and determine various basic indicators based on the raw data; the raw data is data collected from the target vehicle. A semantic indicator construction module is used to encode the basic indicators to obtain the semantic features of the basic indicators; calculate the feature distance based on the semantic features of each basic indicator to determine multiple combinations of basic indicators; each combination of basic indicators includes multiple basic indicators whose feature distance is less than a distance threshold; determine the indicator weight of each basic indicator based on the semantic features of each basic indicator in the same combination of basic indicators; and perform a weighted summation of the basic indicators included in the combination of basic indicators based on the indicator weights of each basic indicator to obtain the semantic indicator corresponding to the combination of basic indicators. The scenario indicator construction module is used to aggregate indicators based on the semantic indicators, determine the combination of semantic indicators, and generate scenario indicators based on the combination of semantic indicators. The scene recognition module is used to extract features based on at least the scene indicators to obtain scene features, and to determine the scene recognition result for the target vehicle based on the scene features.

7. The apparatus according to claim 6, characterized in that, The scenario indicator construction module is used to obtain indicator aggregation examples; the indicator aggregation examples include reference scenario indicators and reference semantic indicator combinations corresponding to the reference scenario indicators. Based on the aforementioned indicator aggregation example, the indicator prediction model aggregates the semantic indicators to determine the combinations of semantic indicators; and generates scene indicators based on the combinations of semantic indicators.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images