A high-risk residual object early warning method for a vehicle based on multi-modal perception

Abstract

The application discloses a kind of based on multimodal perception vehicle high-risk left-over early warning method, it is related to the technical field of in-vehicle safety monitoring, including the image data, point cloud data and environmental temperature data in vehicle are synchronously collected, generate multimodal perception data;According to the suspected left-over area of multimodal perception data extraction, obtain radar reflection characteristics and visual features in suspected left-over area, fusion forms multimodal target feature, by multimodal target feature and pre-stored high-risk item feature library are matched, and the class of left-over is identified;According to the preset basic risk value of left-over of left-over class, combine environmental temperature data to correct basic risk value, generate standardization risk value, according to standardization risk value carries out left-over risk grade judgment.The application can reduce the misjudgment risk caused by single perception mode, improve the identification accuracy of high-risk left-over of vehicle and the scientificity of early warning decision.

Description

Technical Field

[0001] This invention relates to the field of in-vehicle safety monitoring technology, and in particular to a method for early warning of high-risk abandoned objects in vehicles based on multimodal perception. Background Technology

[0002] With the rapid development of intelligent vehicles and in-vehicle sensing technologies, the research focus of vehicle safety protection has gradually expanded from traditional driving safety to in-vehicle safety and safety in various usage scenarios. Existing in-vehicle sensing systems are maturing in areas such as camera technology, millimeter-wave radar sensing, and environmental parameter acquisition, making it possible to comprehensively perceive the status of occupants, objects, and the environment inside the vehicle. Simultaneously, addressing in-vehicle safety issues such as children left behind or dangerous items left behind, related research is gradually incorporating visual perception, target recognition, and environmental monitoring methods to identify and issue warnings for abnormal in-vehicle conditions, thus promoting the development of intelligent in-vehicle monitoring and active safety technologies.

[0003] However, existing technologies still have significant shortcomings in vehicle debris safety monitoring. On the one hand, most solutions rely on single visual information or simple environmental parameters for judgment, which are easily affected by changes in lighting, occlusion, and the similarity of the target's appearance, making it difficult to accurately distinguish between high-risk debris and ordinary items. On the other hand, existing debris identification methods usually focus on the category of the item itself, lacking systematic modeling of the impact of environmental factors, and cannot dynamically assess the actual risk level of the debris based on changes in the in-vehicle environment. In addition, some technologies only provide simple alarm mechanisms, failing to establish a comprehensive risk quantification model and graded early warning strategy, resulting in low generalization and insufficient practicality of the warning results, making it difficult to meet the refined safety management needs in complex in-vehicle scenarios. Summary of the Invention

[0004] In view of the problems existing in the current method for early warning of high-risk debris left by vehicles based on multimodal perception, this invention is proposed. Therefore, the problem to be solved by this invention is how to provide an early warning method for high-risk debris left by vehicles based on multimodal perception.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for early warning of high-risk abandoned objects in vehicles based on multimodal perception, which includes: when a vehicle meets the triggering conditions, simultaneously collecting image data, point cloud data and ambient temperature data inside the vehicle through sensors, processing the image data, point cloud data and ambient temperature data to generate multimodal perception data; Based on multimodal perception data, suspected abandoned object areas are extracted, radar reflection features and visual features in the suspected abandoned object areas are obtained, radar reflection features and visual features are fused to form multimodal target features, and the abandoned object categories are identified by matching the multimodal target features with a pre-stored high-risk item feature database. The basic risk value of the abandoned items is obtained based on the category of abandoned items. The basic risk value is then corrected by combining the environmental temperature data. The corrected risk value is then mapped to generate a standardized risk value. The risk level of the abandoned items is then judged based on the standardized risk value.

[0006] As a preferred embodiment of the high-risk abandoned object warning method for vehicles based on multimodal perception described in this invention, the triggering conditions include: The vehicle is detected to be in a non-driving state. The non-driving state includes at least the vehicle engine being turned off, the vehicle being put into parking mode, and all doors being closed. Based on the in-vehicle sensing results, if it is confirmed that there are no active personnel inside the vehicle, the vehicle is determined to be in an unoccupied state, triggering the activation of sensors, including the in-vehicle camera, millimeter-wave radar, and environmental information acquisition module.

[0007] As a preferred embodiment of the high-risk debris warning method for vehicles based on multimodal perception described in this invention, the generation of multimodal perception data includes: The collected image data, point cloud data, and temperature data are time-aligned to construct a unified perception timeline; Based on the spatial installation relationship of sensors, a spatial mapping model between the camera imaging coordinate system and the millimeter-wave radar detection coordinate system is established to convert point cloud data to a reference coordinate system consistent with image data. The time-aligned image data, point cloud data, and ambient temperature data are correlated and integrated to generate multimodal sensing data.

[0008] As a preferred embodiment of the high-risk abandoned object warning method based on multimodal perception vehicle described in this invention, the step of extracting suspected abandoned object areas based on multimodal perception data includes: Using image data as input, a background update method based on statistical distribution is used to continuously model the in-vehicle scene and construct a dynamic background model of the in-vehicle environment. The image data at the current moment is compared with the dynamic background model inside the vehicle using a pixel-by-pixel difference operation to generate a foreground response map; Connectivity analysis was performed on the foreground response map to extract candidate foreground regions, and these candidate foreground regions were then screened to identify suspected debris areas.

[0009] As a preferred embodiment of the multimodal perception-based high-risk abandoned object early warning method for vehicles described in this invention, the identified abandoned object categories include: Map the spatial extent of the suspected remnant area in the image coordinate system to the corresponding point cloud data, and extract a subset of the point cloud; Radar reflection features of a subset of point clouds are extracted, and visual features are extracted from image data of suspected debris areas using a visual recognition algorithm. By fusing radar reflection features and visual features, multimodal target features are obtained. These multimodal target features are then matched with feature templates in a pre-stored high-risk item feature database to calculate the similarity between the currently suspected abandoned item and the high-risk item, as shown below: ; in, This indicates that the currently suspected remains are related to the first... Similarity between similar items Represents multimodal target features. This indicates the first item in the high-risk item feature database. Feature templates for similar items; The category of items with the highest similarity is selected as the identification result of the category of abandoned items.

[0010] As a preferred embodiment of the high-risk debris warning method for vehicles based on multimodal perception described in this invention, the step of mapping the corrected risk value to generate a standardized risk value includes: Based on the category of abandoned items, a pre-defined category risk mapping relationship is queried to obtain the basic risk value; By introducing current ambient temperature data and correcting the baseline risk value using an environmental risk correction function, the corrected risk value is obtained, expressed as: ; ; in, This represents the environmental risk correction function. This indicates the maximum modulation magnitude of the risk effect caused by temperature. The smoothing coefficient representing the response to risk changes in temperature. Indicates the current ambient temperature. Indicates the reference temperature for risk assessment. This indicates the corrected risk value. Indicates the category of relics The basic risk value; The corrected risk value is then standardized to obtain the standardized risk value, using the following formula: ; in, Represents standardized risk value, This indicates the lowest risk value within the risk reference range. This indicates the highest risk value within the risk reference range.

[0011] As a preferred embodiment of the high-risk abandoned object early warning method based on multimodal perception of vehicles described in this invention, the step of judging the risk level of abandoned objects based on standardized risk values ​​includes: When the standardized risk value is greater than the first preset risk threshold, it is judged as a high-risk level. The vehicle will continuously sound its horn and flash its hazard lights, simultaneously send an emergency notification to the owner's mobile terminal, and upload alarm information to the cloud server. When the standardized risk value is between the first preset risk threshold and the second preset risk threshold, it is determined to be a medium risk level. The vehicle will sound its horn twice, flash its hazard lights multiple times in succession, and push a notification message containing the type of abandoned item, risk level and vehicle status to the owner's mobile terminal. When the standardized risk value is less than the second preset risk threshold, it is determined to be a low-risk level, the vehicle's hazard lights are controlled to flash once, and a prompt message related to the abandoned object is displayed on the vehicle's display screen.

[0012] In a second aspect, the present invention 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 a method for early warning of high-risk abandoned objects from vehicles based on multimodal perception.

[0013] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements the steps of a method for early warning of high-risk abandoned objects from vehicles based on multimodal perception.

[0014] The beneficial effects of this invention are as follows: By constructing multimodal perception data, this invention can effectively improve the integrity and stability of in-vehicle object perception; by combining visual features and radar reflection features to perform multimodal feature fusion and matching identification of suspected objects, it reduces the risk of misjudgment caused by factors such as lighting and occlusion in a single perception method; by introducing ambient temperature to dynamically correct the basic risk of objects, it realizes the quantitative assessment and classification of the danger level of objects, and improves the accuracy of vehicle identification of high-risk objects and the scientific nature of early warning decisions. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a method for early warning of high-risk debris left behind by vehicles based on multimodal perception. Detailed Implementation

[0017] To make the above-mentioned objects, features, and advantages of the present invention more readily understood, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0019] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0020] Reference Figure 1 This is the first embodiment of the present invention, which provides a method for early warning of high-risk debris left behind by vehicles based on multimodal perception, including: S1: When the vehicle meets the triggering conditions, the sensor synchronously collects image data, point cloud data and ambient temperature data inside the vehicle, processes the image data, point cloud data and ambient temperature data to generate multimodal perception data. S2: Extract suspected abandoned object areas based on multimodal perception data, obtain radar reflection features and visual features in the suspected abandoned object areas, fuse radar reflection features and visual features to form multimodal target features, and match the multimodal target features with a pre-stored high-risk item feature database to identify the abandoned object category; S3: Obtain the preset basic risk value of the abandoned items based on the category of abandoned items, correct the basic risk value by combining it with the ambient temperature data, map the corrected risk value to generate a standardized risk value, and judge the risk level of the abandoned items based on the standardized risk value.

[0021] Specifically, when the vehicle is in a pre-set operating condition and the triggering conditions are met, the sensors are activated to collect data synchronously. The sensors include an in-vehicle camera, millimeter-wave radar, and an environmental information acquisition module. The triggering condition is determined by a combination of the vehicle status and the in-vehicle behavior status. The triggering judgment is initiated after the vehicle is detected to have entered a non-driving state. The non-driving state includes at least the vehicle being turned off, the vehicle being in a parking state, and all doors being closed. Further, combined with the in-vehicle perception results, it is confirmed that there are no active personnel in the vehicle, and the vehicle is determined to be in an unoccupied state, triggering the activation of the sensor.

[0022] Image data of the cockpit and occupant activity area are acquired through in-vehicle cameras, point cloud data is acquired by scanning the interior space with millimeter-wave radar, and temperature data of the current environment of the vehicle is collected through an environmental information acquisition module.

[0023] The acquired image data, point cloud data, and temperature data are time-aligned to construct a unified perception timeline, and the image data, point cloud data, and temperature data are rearranged. Spatial registration processing is performed on image data and point cloud data. Based on the spatial installation relationship of the sensors, a spatial mapping model between the camera imaging coordinate system and the millimeter-wave radar detection coordinate system is established to transform the point cloud data from the millimeter-wave radar detection coordinate system to a reference coordinate system consistent with the image data. Image data, point cloud data, and ambient temperature data are correlated and integrated to generate multimodal sensing data; Based on multimodal perception data, image data is used as input to continuously model the in-vehicle scene and construct an in-vehicle dynamic background model. The in-vehicle background model is used to characterize the in-vehicle structure and the state of fixed objects, forming an overall description of the in-vehicle static environment. The in-vehicle dynamic background model is modeled using a background update method based on statistical distribution. By modeling the pixel stability in continuous images, a stable background for the in-vehicle scene is gradually formed, and the in-vehicle dynamic background model is updated when a new image is input.

[0024] The image data at the current perception moment is compared with the in-vehicle dynamic background model by pixel-by-pixel difference operation to obtain the degree of difference between the current image and the in-vehicle dynamic background model in spatial position, and a foreground response map is formed. Connectivity analysis was performed on the foreground response map to extract regions that differed significantly from the in-vehicle dynamic background model as candidate foreground regions. Furthermore, the candidate foreground regions were screened based on the stability of their existence over continuous time. Regions that persisted in multiple adjacent perception moments and whose spatial positions remained largely unchanged were identified as suspected remnant regions.

[0025] Based on spatial registration, the spatial range of the suspected remnant area in the image coordinate system is mapped to the corresponding point cloud data, and a subset of the point cloud located within that spatial range is extracted. For a subset of point clouds, radar reflection characteristics reflecting the spatial scale, point cloud distribution density, and reflection intensity of the target are obtained to describe the physical properties of suspected remnants at the radar sensing level. Simultaneously, visual recognition processing is performed on the image data of the suspected remnant area, and visual features are extracted from the image data within the area to form visual features that characterize the target's appearance, contour structure, and surface texture. Radar reflection features are fused with visual features to obtain multimodal target features. The visual recognition algorithm uses a target recognition method based on a learning model to identify the target category in areas of suspected debris in an image; The multimodal target features are matched with a pre-established high-risk item feature database. This database includes multimodal feature templates for each category of high-risk items. The similarity between the current target feature and each feature template is calculated and expressed as: ; in, This indicates that the currently suspected remains are related to the first... Similarity between similar items Represents multimodal target features. This indicates the first item in the high-risk item feature database. Feature templates for similar items.

[0026] The high-risk item feature database was established by centrally collecting samples of various typical in-vehicle items, covering at least common personal items and items with potential safety hazards. The multimodal feature template for each item category was obtained through statistical fusion of the sample data, used to characterize the common features of that item category at both the visual and radar perception levels.

[0027] Based on the similarity calculation results between the current suspected abandoned items and high-risk items, the category of the item with the highest similarity to the current suspected abandoned items is determined as the identification result of the abandoned items, thus completing the identification of the category of abandoned items in the vehicle; After completing the identification of abandoned items categories, the basic risk value of each abandoned item category is obtained based on the mapping relationship between abandoned item categories and preset risks. The basic risk value is pre-established offline during the system deployment phase. It is determined by the mapping relationship between the category of abandoned items and the risk level. The basis for setting the risk value includes the physical characteristics of the item, its potential hazards, and the safety consequences that may be caused in the closed vehicle environment. The mapping relationship comes from a comprehensive analysis of historical safety incident statistics and safety assessment experience.

[0028] By incorporating ambient temperature data corresponding to the current sensing moment, the baseline risk value is corrected. The ambient temperature data is then correlated with the risk-sensitive characteristics of the debris, constructing an environmental risk correction function, expressed as: ; ; in, This represents the environmental risk correction function. This indicates the maximum modulation magnitude of the risk effect caused by temperature. The smoothing coefficient representing the response to risk changes in temperature. Indicates the current ambient temperature. Indicates the reference temperature for risk assessment. This indicates the corrected risk value. Indicates the category of relics The basic risk value; After obtaining the corrected risk value, the corrected risk value is mapped using a preset risk reference range as a benchmark. This mapping is then proportional to the corrected risk value, forming a standardized risk value. The standardized risk value reflects the relative danger level of the current legacy, and is expressed as: ; in, Represents standardized risk value, This indicates the lowest risk value within the risk reference range. This indicates the highest risk value within the risk reference range; The risk level of abandoned items is determined based on standardized risk values. The risk level includes three states: low risk, medium risk, and high risk, which are used to reflect the overall degree of danger of the abandoned items under vehicle and environmental conditions.

[0029] The risk level is determined based on the value of the final risk indicator. When the risk indicator is greater than the first preset risk threshold, the risk level is determined to be high risk. The system continuously controls the vehicle to sound its horn and keep the hazard lights on to create a warning effect on site. At the same time, an emergency notification message is sent to the vehicle owner's mobile terminal, and the alarm data is uploaded to the cloud server. When the risk indicator is between the first preset risk threshold and the second preset risk threshold, the risk level is determined to be medium risk. The vehicle is controlled to sound its horn twice and drive the vehicle's hazard lights to flash continuously multiple times. At the same time, the vehicle communication module pushes a notification to the owner's mobile terminal containing information on the type of abandoned item, the risk level, and the current vehicle status. When the risk index is less than the second preset risk threshold, the risk level is determined to be low risk, indicating that the danger level of the abandoned object under the current environmental conditions is low. Only basic prompts are needed to remind the people in the vehicle to pay attention. The system controls the vehicle to execute a low-level warning response, and sends a control command to the body control module to make the vehicle's hazard lights flash once. At the same time, the system outputs prompt information related to the type and location of the abandoned object on the vehicle display screen.

[0030] This embodiment also provides a computer device applicable to a method for early warning of high-risk abandoned objects from vehicles based on multimodal perception, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement all or part of the steps of the method described in the above embodiments of the present invention.

[0031] This embodiment also provides a storage medium storing a computer program thereon. When the computer program is executed by a processor, it performs the method in any optional implementation of the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0032] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0033] In summary, this invention effectively improves the integrity and stability of in-vehicle object detection by constructing multimodal perception data. Based on this, it combines visual and radar reflection features to perform multimodal feature fusion and matching for suspected objects, significantly reducing the risk of misjudgment caused by factors such as lighting and occlusion in single-sensor methods. By introducing ambient temperature to dynamically correct the basic risk of objects and mapping the risk quantity to standardized risk values, it achieves quantitative assessment and grading of the danger level of objects, improving the accuracy of vehicle identification of high-risk objects and the scientific nature of early warning decisions.

[0034] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for early warning of high-risk abandoned items on vehicles based on multimodal perception, characterized in that: include, When the vehicle meets the triggering conditions, the sensor synchronously collects image data, point cloud data and ambient temperature data inside the vehicle, processes the image data, point cloud data and ambient temperature data to generate multimodal perception data; Based on multimodal perception data, suspected abandoned object areas are extracted, radar reflection features and visual features in the suspected abandoned object areas are obtained, radar reflection features and visual features are fused to form multimodal target features, and the abandoned object categories are identified by matching the multimodal target features with a pre-stored high-risk item feature database. The basic risk value of the abandoned items is obtained based on the category of abandoned items. The basic risk value is then corrected by combining the environmental temperature data. The corrected risk value is then mapped to generate a standardized risk value. The risk level of the abandoned items is then judged based on the standardized risk value.

2. The method for early warning of high-risk abandoned objects on vehicles based on multimodal perception as described in claim 1, characterized in that: The triggering conditions include: The vehicle is detected to be in a non-driving state. The non-driving state includes at least the vehicle being turned off, the vehicle being put into parking mode, and all doors being closed. Based on the in-vehicle sensing results, if it is confirmed that there are no active personnel inside the vehicle, the vehicle is determined to be in an unoccupied state, triggering the activation of sensors, including the in-vehicle camera, millimeter-wave radar, and environmental information acquisition module.

3. The method for early warning of high-risk abandoned objects on vehicles based on multimodal perception as described in claim 1, characterized in that: The generated multimodal sensing data includes: Time alignment processing is performed on the collected image data, point cloud data, and temperature data to construct a unified perception timeline; Based on the spatial installation relationship of sensors, a spatial mapping model between the camera imaging coordinate system and the millimeter-wave radar detection coordinate system is established to convert point cloud data to a reference coordinate system consistent with image data. The time-aligned image data, point cloud data, and ambient temperature data are correlated and integrated to generate multimodal sensing data.

4. The method for early warning of high-risk abandoned objects on vehicles based on multimodal perception as described in claim 1, characterized in that: The extraction of suspected remnant areas based on multimodal sensing data includes: Using image data as input, a background update method based on statistical distribution is used to continuously model the in-vehicle scene and construct an in-vehicle dynamic background model. The image data at the current moment is compared with the dynamic background model inside the vehicle using a pixel-by-pixel difference operation to generate a foreground response map; Connectivity analysis was performed on the foreground response map to extract candidate foreground regions, and these candidate foreground regions were then screened to identify suspected debris areas.

5. The method for early warning of high-risk abandoned objects on vehicles based on multimodal perception as described in claim 4, characterized in that: The categories of identified abandoned items include: Map the spatial extent of the suspected remnant area in the image coordinate system to the corresponding point cloud data, and extract a subset of the point cloud; Radar reflection features of a subset of point clouds are extracted, and visual features are extracted from image data of suspected debris areas using a visual recognition algorithm. By fusing radar reflection features and visual features, multimodal target features are obtained. These multimodal target features are then matched with feature templates in a pre-stored high-risk item feature database to calculate the similarity between the currently suspected abandoned item and the high-risk item, as shown below: ； in, This indicates that the currently suspected remains are related to the first... Similarity between similar items Represents multimodal target features. This indicates the first item in the high-risk item feature database. Feature templates for similar items; The category of items with the highest similarity is selected as the identification result of the category of abandoned items.

6. The method for early warning of high-risk abandoned objects on vehicles based on multimodal perception as described in claim 1, characterized in that: The step of mapping the corrected risk value to generate a standardized risk value includes: Based on the category of abandoned items, a pre-defined category risk mapping relationship is queried to obtain the basic risk value; By introducing current ambient temperature data and correcting the baseline risk value using an environmental risk correction function, the corrected risk value is obtained, expressed as: ； ； in, This represents the environmental risk correction function. This indicates the maximum modulation magnitude of the risk effect caused by temperature. The smoothing coefficient representing the response to risk changes in temperature. Indicates the current ambient temperature. Indicates the reference temperature for risk assessment. This indicates the corrected risk value. Indicates the category of relics The basic risk value; The corrected risk value is then standardized to obtain the standardized risk value, using the following formula: ； in, Represents standardized risk value, This indicates the lowest risk value within the risk reference range. This indicates the highest risk value within the risk reference range.

7. The method for early warning of high-risk abandoned objects on vehicles based on multimodal perception as described in claim 6, characterized in that: The determination of the risk level of legacy items based on standardized risk values ​​includes: When the standardized risk value is greater than the first preset risk threshold, it is judged as a high-risk level. The vehicle will continuously sound its horn and flash its hazard lights, simultaneously send an emergency notification to the owner's mobile terminal, and upload alarm information to the cloud server. When the standardized risk value is between the first preset risk threshold and the second preset risk threshold, it is determined to be a medium risk level. The vehicle will sound its horn twice, flash its hazard lights multiple times in succession, and push a notification message containing the type of abandoned item, risk level and vehicle status to the owner's mobile terminal. When the standardized risk value is less than the second preset risk threshold, it is determined to be a low-risk level, the vehicle's hazard lights are controlled to flash once, and a prompt message related to the abandoned object is displayed on the vehicle's display screen.

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 for early warning of high-risk abandoned objects on vehicles based on multimodal perception, as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for early warning of high-risk abandoned objects on vehicles based on multimodal perception as described in any one of claims 1 to 7.

Images