Theme-driven multi-camera self-adaptive split-screen shooting method and system for vehicle-mounted camera

Abstract

The application discloses a theme-driven vehicle-mounted multi-camera adaptive split-screen shooting method and system, relates to the split-screen shooting field, and comprises the following steps: S1, acquiring data; S2, determining a target subject and a visual angle based on a theme; S3, scoring, sorting and role assigning of each vehicle-mounted camera; S4, generating an adaptive split-screen template based on screening; S5, acquiring a frame set corresponding to effective images; S6, preprocessing and split-screen synthesizing of the frame set of the effective images; and S7, quality evaluation of output split-screen results, and iterative optimization of shooting control strategies based on evaluation data. The application introduces theme parameters, camera capability modeling, adaptive camera selection and abandonment, synchronous acquisition and cross-time key frame joint arrangement, and work data training closed loop, and solves the problems that the multi-camera result generation in the prior art is disconnected with a specific vehicle service scene, is difficult to adapt to camera configuration changes, is difficult to completely retain key processes, and is difficult to continuously optimize control strategies.

Description

Technical Field

[0001] This invention relates to the field of storyboard shooting, and more particularly to a subject-driven vehicle-mounted multi-camera adaptive storyboard shooting method and a subject-driven vehicle-mounted multi-camera adaptive storyboard shooting system. Background Technology

[0002] With the development of intelligent cockpits, driver assistance systems, and in-vehicle imaging applications, vehicles are typically equipped with various image acquisition devices, including front-view cameras, rear-view cameras, side-view cameras, surround-view cameras, driver monitoring cameras, and occupant monitoring cameras. Current technologies generally use these cameras for single scenarios such as parking, blind spot assistance, driver status recognition, driving recording, or entertainment recording, rarely considering the unified arrangement and structured output of multi-camera acquisition results from the perspectives of "vehicle services," "driving services," and "vehicle safety services."

[0003] In scenarios such as driving process recording, parking process review, dangerous target retention, and driver status correlation display, existing multi-camera solutions typically employ fixed tile preview, single-channel switching viewing, or post-event manual selection and stitching. These methods struggle to automatically select the most relevant footage from multiple heterogeneous cameras based on the current task theme, and also fail to produce output results with clear priorities, complete structure, and reviewability.

[0004] Furthermore, there are significant differences in camera configurations across different vehicle models in terms of quantity, installation location, field of view, resolution, frame rate, low-light capability, dynamic range, and availability. Existing solutions generally rely on fixed camera combinations and fixed display templates. When a vehicle lacks certain cameras, some cameras temporarily malfunction, the image is severely obstructed, or the performance is insufficient, the output results are prone to problems such as missing key subjects, repetitive viewpoints, structural imbalances, or generation failures, making it impossible to achieve stable adaptation across vehicle models and configurations.

[0005] Furthermore, most existing multi-camera shooting solutions only process synchronized footage at the same moment, and their representation of driving behavior and external events remains at the level of instantaneous static stitching, lacking a mechanism to organize different time frames around the event trigger point. For example, in scenarios such as meeting oncoming traffic on a narrow road, reversing into a parking space, emergency braking, and driver fatigue warning, the truly valuable information is often distributed across different time periods before, during, and after the event. If only the same frame image is retained, it is impossible to fully reflect the dynamic relationship between the vehicle, the environment, and the driver.

[0006] On the other hand, existing solutions lack quantifiable assessments of the quality of generated results and also lack a closed-loop mechanism for using historical work results to train models and optimize shooting control. In other words, existing systems struggle to continuously improve camera scheduling strategies, keyframe extraction strategies, and storyboarding strategies based on evaluation metrics such as subject matching degree, subject coverage, key event retention rate, viewpoint redundancy, and synchronization integrity.

[0007] Therefore, existing technologies suffer from at least the following problems: First, they lack a multi-camera collaborative control and result generation mechanism for subject-specific tasks; second, they lack an adaptive camera selection mechanism for changes in heterogeneous camera configurations; third, they lack a segmentation organization mechanism for joint arrangement of simultaneous frames and cross-time frames; and fourth, they lack a continuous optimization mechanism for the shooting control model based on the final product. These problems make it difficult for existing solutions to stably serve scenarios such as driving recording, external target retention, in-vehicle state correlation display, and vehicle safety incident review. Summary of the Invention

[0008] The purpose of this invention is to provide a subject-driven vehicle-mounted multi-camera adaptive split-scene shooting method and a subject-driven vehicle-mounted multi-camera adaptive split-scene shooting system, thereby solving at least one of a number of technical problems.

[0009] Key technical challenges include: the disconnect between multi-camera results generation and specific vehicle service scenarios; difficulty in adapting to changes in camera configuration; difficulty in fully preserving key processes; and difficulty in continuously optimizing control strategies.

[0010] This invention provides the following solution:

[0011] According to a first aspect of the present invention, a subject-driven adaptive split-scene shooting method for in-vehicle multi-camera systems is provided, comprising the following steps:

[0012] S1. Obtain the topic parameters, vehicle status data, and camera capability data of each vehicle camera for the current task;

[0013] S2. Construct a topic-perspective requirement model based on topic parameters to determine the target subject, essential perspectives, and discardable perspectives for the current task.

[0014] S3. Based on the camera capability data and subject-viewpoint demand model of each vehicle camera, score the contribution of each vehicle camera, prioritize and assign roles, select high-contribution cameras and remove low-contribution cameras to select usable cameras.

[0015] S4. Based on the number of available cameras after filtering, the role allocation results and the current theme parameters, dynamically generate an adapted storyboard template. The storyboard template includes: storyboard areas with different priorities; it also includes, according to the area, position and level of each area in the storyboard template, prioritizing the display of the main storyboard that is most relevant to the theme, and supplementing the auxiliary storyboard with spatial relationships, driver status, vehicle operation or environmental context.

[0016] S5. Based on the task requirements of the corresponding theme parameters, selectively execute the synchronous acquisition mode or the cross-temporal keyframe joint arrangement mode to obtain the frame set of the corresponding valid images.

[0017] S6. Preprocess and synthesize the frame set of effective images, and output the vehicle-mounted storyboard result with theme matching.

[0018] S7. Evaluate the quality of the output storyboard results and iteratively optimize the shooting control strategy based on the evaluation data.

[0019] Furthermore, the topic parameters include:

[0020] At least one of the following must be included: theme type, target subject type, target event type, time window, maximum number of storyboards, output style, required viewpoints, optional viewpoints, security priority, and triggering method.

[0021] The topic parameters can be entered manually or generated automatically.

[0022] Theme parameters are automatically generated based on vehicle status;

[0023] It also includes camera capability data: physical attributes, performance indicators, status information, and functional attributes;

[0024] Physical attributes include: installation location, orientation, field of view, resolution, and frame rate;

[0025] Performance metrics include: low-light capability, dynamic range, and imaging stability;

[0026] Status information includes: online status, fault status, obstruction status, and synchronization status;

[0027] Functional attributes include: object detection, face recognition, and behavior recognition capability tags.

[0028] Furthermore, in step S3, the contribution score is calculated as follows:

[0029] Based on the theme relevance score R, the image quality score Q, and the perspective synergy and complementarity score C, the weighted comprehensive score S is calculated as follows: S = α × R + β × Q + γ × C.

[0030] Where α, β, and γ are preset weighting coefficients, and R, Q, and C are all normalized to the 0-100 score range;

[0031] The topic relevance score R is calculated from the target subject pixel area ratio, topic feature hit rate, and event window matching degree.

[0032] The image quality score Q is calculated from the percentage of effective clear pixels, the percentage of exposed pixels, and the occlusion area ratio.

[0033] The viewpoint coordination complementarity score C is calculated from the newly added visible area ratio, cross-camera timestamp difference, and the proportion of viewpoint redundancy area.

[0034] It also includes a dynamic weight adjustment mechanism: when the target subject pixel area ratio, topic feature hit rate, and event window matching degree meet the preset high-quality conditions, the weight α corresponding to the topic relevance score is increased.

[0035] When the hit rate of the subject feature is lower than the preset hit threshold, or / and the effective proportion of the target subject is lower than the preset proportion threshold, the weight α is reduced and the corresponding camera is determined to be a low contribution camera in this round.

[0036] Furthermore, the theme-perspective requirement model includes establishing mapping rules between themes, main characters, and perspective sets:

[0037] The main subject is dynamically switched according to different theme types. The main subjects include: driver, external target, road risk area, parking area, and vehicle posture.

[0038] The matching process ensures that the viewpoint is retained or discarded, enabling adaptive viewpoint adaptation for different scenarios.

[0039] Furthermore, in step S4, a dynamically generated storyboard template is created, including an adaptive degradation mechanism:

[0040] When the vehicle-mounted camera configuration is complete and the number of available cameras is sufficient, a complete template including the main storyboard, auxiliary storyboard, supplementary storyboard, and panoramic storyboard is generated.

[0041] When there are missing, faulty, obstructed, or insufficient cameras, the system automatically reduces the number of auxiliary shots, expands the display area of ​​the main shot, removes redundant shots, and generates a simplified adaptation template.

[0042] For different vehicle models with heterogeneous camera configurations, the complexity of the storyboard template is dynamically adjusted to maintain the consistency and integrity of the theme display.

[0043] Furthermore, the execution logic of the synchronous acquisition mode includes:

[0044] Time alignment of image frames from multiple cameras is performed using a unified vehicle time reference.

[0045] Among them, a preset time deviation threshold is set to filter and obtain a set of image frames that meet the synchronization conditions at the same time.

[0046] It also includes: replacing camera frames that exceed the time deviation threshold or fail with the nearest available frame or enabling backup view compensation, and marking non-strict synchronization states.

[0047] Among them, the synchronous acquisition mode is primarily suitable for scenarios involving instantaneous status recording.

[0048] Furthermore, the execution logic of the cross-temporal keyframe joint orchestration mode includes:

[0049] Determine a preset time window based on the target event, and extract candidate image frames from each camera within the time window;

[0050] Candidate image frames are scored and filtered based on topic matching degree, image quality, event criticality, and viewpoint complementarity, and key frames before, during, and after the event are selected respectively.

[0051] Keyframes at different time intervals are mapped to corresponding areas of the storyboard template to form a storyboard sequence that can reconstruct the complete event process.

[0052] Furthermore, the preprocessing in step S6 includes:

[0053] At least one of the following: distortion correction, target area cropping, brightness adjustment, noise reduction and enhancement, boundary generation, identifier overlay, and style rendering;

[0054] Style rendering includes differentiated visual outputs such as documentary, abstract, and comic styles, while maintaining the clarity of the target subject at the preset resolution during the style rendering process.

[0055] Furthermore, in step S7, the quality assessment employs multi-dimensional quantitative indicators:

[0056] Quantitative metrics include: topic matching score, subject coverage, key event retention rate, perspective redundancy, clarity score, synchronization completeness, and user operation feedback metrics.

[0057] The evaluation data, subject parameters, camera data, and frame selection data of each storyboard result are associated and stored as training samples;

[0058] The camera scoring parameters, keyframe extraction parameters, and storyboard template parameters are iteratively updated based on training samples to achieve continuous optimization of the shooting control strategy.

[0059] According to a second aspect of the present invention, a subject-driven vehicle-mounted multi-camera adaptive shot-capture system is provided for executing a subject-driven vehicle-mounted multi-camera adaptive shot-capture method. The subject-driven vehicle-mounted multi-camera adaptive shot-capture system includes:

[0060] The topic parameter parsing module is used to receive or automatically generate topic parameters for the current task;

[0061] The vehicle status acquisition module is used to acquire vehicle status, environmental and driving event data;

[0062] The camera capability modeling module is used to build and store the structured capability data of each vehicle camera;

[0063] The topic-perspective requirements modeling module is used to determine the target subject and perspective requirements set based on topic parameters;

[0064] The camera scoring and low-contribution screening module is used to score, sort, and remove low-contribution devices from cameras.

[0065] The storyboard template generation module is used to dynamically generate and adaptively downgrade and adjust the storyboard layout template.

[0066] The synchronous acquisition module and the time-series keyframe extraction module are used to execute the synchronous acquisition mode and the cross-time-series keyframe orchestration mode, respectively.

[0067] The storyboard compositing module is used to complete image preprocessing and compositing output of storyboard results;

[0068] The work quality assessment module and the shooting control model training module are used to complete the closed loop of result quantitative evaluation and strategy iterative optimization;

[0069] Each module is deployed in the vehicle central control terminal, cockpit computing platform, or vehicle domain controller.

[0070] The above solution achieves the following beneficial technical effects:

[0071] This invention addresses the problems in existing technologies, such as the disconnect between multi-camera result generation and specific vehicle service scenarios, difficulty in adapting to changes in camera configuration, difficulty in fully preserving key processes, and difficulty in continuously optimizing control strategies, by introducing subject parameters, camera capability modeling, adaptive camera selection and discarding, synchronous acquisition and cross-temporal keyframe joint arrangement, and closed-loop training of work data.

[0072] This invention automatically identifies the highlighted target subject based on the current theme, forming a storyboard result that is clear in its primary and secondary elements, structurally complete, and recapable; it automatically determines the main storyboard, auxiliary storyboard, supplementary storyboard, and low-contribution storyboard based on differences in camera capabilities; it switches between synchronous acquisition at the same time and keyframe extraction across time windows; it downgrades and reconstructs the storyboard layout when a camera is missing, its performance degrades, or it experiences a temporary malfunction; and it trains a shooting control model based on the results of historical works to improve the quality of subsequent storyboard results.

[0073] This invention focuses on driving safety, vehicle services, and scene recording, automatically increasing the coverage and display priority of target subjects in the results, and improving the matching degree between the storyboard results and the actual tasks.

[0074] This invention can perform capability evaluation, priority ranking, and low-contribution screening of multiple cameras under different vehicle models, different numbers of cameras, and different equipment capabilities, thereby improving the success rate of cross-vehicle storyboard generation and reducing the output failure rate due to missing or abnormal cameras.

[0075] This invention can ensure the temporal consistency of multiple images in synchronous acquisition mode and retain key frames before, during and after an event in cross-time sequence arrangement mode, thereby improving the completeness of the review of driving process and safety events.

[0076] This invention can dynamically adjust the main subject object according to the topic parameters. It can take the driver, external targets, road risks, parking areas or vehicle posture as the main subject, thereby improving the task adaptability of the output results.

[0077] This invention can use the results of work quality assessment to train the shooting control model in reverse, continuously improving the theme matching score, subject integrity score and key event retention rate, forming a technological closed loop of sustainable optimization.

[0078] At the user interaction level, the system can automatically complete multi-camera scheduling, keyframe filtering, storyboard synthesis, and result output through a single theme selection and trigger operation, thereby reducing manual switching, filtering, and editing operations and improving the system's usability and practical value. Attached Figure Description

[0079] Figure 1 This is a flowchart of a subject-driven, vehicle-mounted multi-camera adaptive split-scene shooting method provided by one or more embodiments of the present invention.

[0080] Figure 2 This is a structural diagram of a subject-driven vehicle-mounted multi-camera adaptive split-scene shooting system provided by one or more embodiments of the present invention. Detailed Implementation

[0081] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0082] Figure 1 This is a flowchart of a subject-driven, vehicle-mounted multi-camera adaptive split-scene shooting method provided by one or more embodiments of the present invention.

[0083] Example 1, such as Figure 1 The subject-driven adaptive split-scene shooting method for in-vehicle multi-camera systems, as shown, includes the following steps:

[0084] S1. Obtain the topic parameters, vehicle status data, and camera capability data of each vehicle camera for the current task;

[0085] S2. Construct a topic-perspective requirement model based on topic parameters to determine the target subject, essential perspectives, and discardable perspectives for the current task.

[0086] S3. Based on the camera capability data and subject-viewpoint demand model of each vehicle camera, score the contribution of each vehicle camera, prioritize and assign roles, select high-contribution cameras and remove low-contribution cameras to determine the usable cameras.

[0087] S4. Based on the number of available cameras after filtering, the role allocation results and the current topic parameters, dynamically generate an adapted storyboard template. The storyboard template includes storyboard areas with different primary and secondary roles.

[0088] S5. Based on the task requirements of the corresponding theme parameters, selectively execute the synchronous acquisition mode or the cross-temporal keyframe joint arrangement mode to obtain the frame set of the corresponding valid images.

[0089] S6. Preprocess and synthesize the frame set of effective images, and output the vehicle-mounted storyboard result with theme matching.

[0090] S7. Evaluate the quality of the output storyboard results, and iteratively optimize the shooting control strategy based on the evaluation data through closed-loop optimization.

[0091] In this embodiment, the topic parameters include:

[0092] At least one of the following must be included: theme type, target subject type, target event type, time window, maximum number of storyboards, output style, required viewpoints, optional viewpoints, security priority, and triggering method.

[0093] The topic parameters can be entered manually or generated automatically.

[0094] Theme parameters are automatically generated based on vehicle status;

[0095] It also includes camera capability data: physical attributes, performance indicators, status information, and functional attributes;

[0096] Physical attributes include: installation location, orientation, field of view, resolution, and frame rate;

[0097] Performance metrics include: low-light capability, dynamic range, and imaging stability;

[0098] Status information includes: online status, fault status, obstruction status, and synchronization status;

[0099] Functional attributes include: object detection, face recognition, and behavior recognition capability tags.

[0100] In this embodiment, the calculation method for the contribution score in step S3 includes:

[0101] Based on the theme relevance score R, the image quality score Q, and the perspective synergy and complementarity score C, the weighted comprehensive score S is calculated as follows: S = α × R + β × Q + γ × C.

[0102] Where α, β, and γ are preset weighting coefficients, and R, Q, and C are all normalized to the 0-100 score range;

[0103] The topic relevance score R is calculated from the target subject pixel area ratio, topic feature hit rate, and event window matching degree.

[0104] The image quality score Q is calculated from the percentage of effective clear pixels, the percentage of exposed pixels, and the occlusion area ratio.

[0105] The viewpoint coordination complementarity score C is calculated from the newly added visible area ratio, cross-camera timestamp difference, and the proportion of viewpoint redundancy area.

[0106] It also includes a dynamic weight adjustment mechanism: when the target subject pixel area ratio, topic feature hit rate, and event window matching degree meet the preset high-quality conditions, the weight α corresponding to the topic relevance score is increased.

[0107] When the hit rate of the subject feature is lower than the preset hit threshold, or / and the effective proportion of the target subject is lower than the preset proportion threshold, the weight α is reduced and the corresponding camera is determined to be a low contribution camera in this round.

[0108] In this embodiment, the theme-perspective requirement model includes establishing mapping rules between the theme, the main character object, and the set of perspectives:

[0109] The main subject is dynamically switched according to different theme types. The main subjects include: driver, external target, road risk area, parking area, and vehicle posture.

[0110] The matching process ensures that the viewpoint is retained or discarded, enabling adaptive viewpoint adaptation for different scenarios.

[0111] In this embodiment, when the vehicle-mounted camera configuration is complete and the number of available cameras is sufficient, a complete template including the main split shot, auxiliary split shot, supplementary split shot, and panoramic split shot is generated.

[0112] When there are missing, faulty, obstructed, or insufficient cameras, the system automatically reduces the number of auxiliary shots, expands the display area of ​​the main shot, removes redundant shots, and generates a simplified adaptation template.

[0113] For different vehicle models with heterogeneous camera configurations, the complexity of the storyboard template is dynamically adjusted to maintain the consistency and integrity of the theme display.

[0114] In this embodiment, the execution logic of the synchronous acquisition mode includes:

[0115] Time alignment of image frames from multiple cameras is performed using a unified vehicle time reference.

[0116] Among them, a preset time deviation threshold is set to filter and obtain a set of image frames that meet the synchronization conditions at the same time.

[0117] It also includes: replacing camera frames that exceed the time deviation threshold or fail with the nearest available frame or enabling backup view compensation, and marking non-strict synchronization states.

[0118] Among them, the synchronous acquisition mode is primarily suitable for scenarios involving instantaneous status recording.

[0119] In this embodiment, the execution logic of the cross-temporal keyframe joint orchestration mode includes:

[0120] Determine a preset time window based on the target event, and extract candidate image frames from each camera within the time window;

[0121] Candidate image frames are scored and filtered based on topic matching degree, image quality, event criticality, and viewpoint complementarity, and key frames before, during, and after the event are selected respectively.

[0122] Keyframes at different time intervals are mapped to corresponding areas of the storyboard template to form a storyboard sequence that can reconstruct the complete event process.

[0123] In this embodiment, the preprocessing in step S6 includes:

[0124] At least one of the following: distortion correction, target area cropping, brightness adjustment, noise reduction and enhancement, boundary generation, identifier overlay, and style rendering;

[0125] Style rendering includes differentiated visual outputs such as documentary, abstract, and comic styles, while maintaining the clarity of the target subject at the preset resolution during the style rendering process.

[0126] In this embodiment, the quality assessment in step S7 uses multi-dimensional quantitative indicators:

[0127] Quantitative metrics include: topic matching score, subject coverage, key event retention rate, perspective redundancy, clarity score, synchronization completeness, and user operation feedback metrics.

[0128] The evaluation data, subject parameters, camera data, and frame selection data of each storyboard result are associated and stored as training samples;

[0129] The camera scoring parameters, keyframe extraction parameters, and storyboard template parameters are iteratively updated based on training samples to achieve continuous optimization of the shooting control strategy.

[0130] Figure 2 This is a structural diagram of a subject-driven vehicle-mounted multi-camera adaptive split-scene shooting system provided by one or more embodiments of the present invention.

[0131] Example 2, as Figure 2 The subject-driven adaptive multi-camera split-scene shooting system shown is used to execute the subject-driven adaptive multi-camera split-scene shooting method. The subject-driven adaptive multi-camera split-scene shooting system includes:

[0132] The topic parameter parsing module is used to receive or automatically generate topic parameters for the current task;

[0133] The vehicle status acquisition module is used to acquire vehicle status, environmental and driving event data;

[0134] The camera capability modeling module is used to build and store the structured capability data of each vehicle camera;

[0135] The topic-perspective requirements modeling module is used to determine the target subject and perspective requirements set based on topic parameters;

[0136] The camera scoring and low-contribution screening module is used to score, sort, and remove low-contribution devices from cameras.

[0137] The storyboard template generation module is used to dynamically generate and adaptively downgrade and adjust the storyboard layout template.

[0138] The synchronous acquisition module and the time-series keyframe extraction module are used to execute the synchronous acquisition mode and the cross-time-series keyframe orchestration mode, respectively.

[0139] The storyboard compositing module is used to complete image preprocessing and compositing output of storyboard results;

[0140] The work quality assessment module and the shooting control model training module are used to complete the closed loop of result quantitative evaluation and strategy iterative optimization;

[0141] Each module is deployed in the vehicle central control terminal, cockpit computing platform, or vehicle domain controller.

[0142] It is worth noting that although this system / device only discloses the above-mentioned modules / units, it does not mean that this system / device is limited to the above-mentioned basic functional modules. On the contrary, what this invention intends to express is that, based on the above-mentioned basic functional modules, those skilled in the art can add one or more functional modules in combination with the prior art to form an infinite number of embodiments or technical solutions. That is to say, this system is open rather than closed. It cannot be assumed that the scope of protection of the claims of this invention is limited to the above-disclosed basic functional modules just because this embodiment only discloses a few basic functional modules.

[0143] Example 3, in a specific embodiment, discloses a system architecture for theme-driven adaptive multi-camera storyboard shooting in vehicles: a theme parameter parsing module, a vehicle state acquisition module, a camera capability modeling module, a theme-viewpoint requirement modeling module, a camera scoring and low-contribution screening module, a storyboard template generation module, a synchronous acquisition module, a temporal keyframe extraction module, a storyboard compositing module, a work quality evaluation module, and a shooting control model training module. These modules can be deployed in an in-vehicle central control terminal, a cockpit computing platform, or a domain controller communicating with the in-vehicle cameras.

[0144] The system comprises the following modules: a topic parameter parsing module for receiving user-input topics, preset scripts, or task topics automatically generated by the system based on vehicle status; a vehicle status acquisition module for acquiring vehicle and environmental information such as vehicle speed, steering angle, gear position, braking status, lighting status, radar targets, driver status, and ADAS event status; a camera capability modeling module for constructing capability labels for each camera; a topic-viewpoint requirement modeling module for generating target subjects and required viewpoint sets based on topic parameters; a camera scoring and low-contribution filtering module for selecting the most suitable camera combination; a storyboard template generation module for generating storyboard layouts with clear hierarchy; a synchronous acquisition module and a temporal keyframe extraction module for processing same-frame mode and cross-temporal mode respectively; a storyboard compositing module for outputting the final storyboard image or storyboard sequence; and a work quality evaluation module and a shooting control model training module for forming a closed loop for subsequent optimization.

[0145] Example 4, (a) Topic Parameter Driven Mechanism

[0146] This invention uses subject parameters as the starting point for shooting control. Subject parameters include at least one or more of the following: subject type, target subject type, target event type, time window, maximum number of scenes, output style, required viewpoint, optional viewpoint, safety priority, and triggering method. Subject parameters can be selected and input by the user on the in-vehicle terminal, or automatically inferred and generated by the system based on the current vehicle status (as shown in Table 1).

[0147] Table 1

[0148] Parameter type illustrate Example value Theme type Task Scenario Classification Driving recorder, parking debriefing, hazardous target retention, driver status recorder Target subject type The key points to be emphasized Driver, external targets, road risks, parking area, vehicle posture Target event type Event types that trigger storyboard generation Fatigue driving warning, passing oncoming vehicles in narrow roads, reversing into a parking space, emergency braking Time window The time range before and after the event From 2 seconds before the event to 3 seconds after the event Maximum number of storyboards The maximum number of storyboards allowed in the storyboard results 3, 4 or 5 Result output style Result presentation Documentary style, summary style, cartoon style Must retain perspective The perspective that must be included in the generated results Driver monitoring, forward-looking road A perspective that can be discarded Perspectives that can be prioritized for filtering when resources are insufficient Rear view, partial side view Security Priority Determine the level of recording and retention strategy High, Medium, Low Triggering method How to start a storyboard task Manual trigger, automatic trigger of ADAS events

[0149] Example 5, (II) Camera Capability Modeling and Adaptive Camera Selection Mechanism

[0150] To adapt to the heterogeneous camera configurations of different vehicles, this invention establishes a capability description set for each camera. The capability description set can be divided according to physical attributes, performance indicators, status information, and functional attributes (as shown in Table 2).

[0151] Table 2

[0152] Capability Dimension Specific attributes illustrate Physical properties Installation location, orientation, horizontal field of view, vertical field of view, resolution, frame rate Used to determine coverage and basic imaging capabilities Performance indicators Low-light capability, dynamic range, distortion parameters, imaging stability Used to evaluate usability in different environments Status information Online status, fault status, occlusion detection results, synchronization status Used to identify whether someone is currently suitable to participate in storyboard generation. Functional attributes Image pattern recognition, object detection, face recognition, and behavior recognition capabilities. Used to determine its auxiliary value to a specific topic.

[0153] Example 6: In one possible implementation, the set of camera capability descriptions can be stored and exchanged using structured data objects, such as JSON or XML formats. The following JSON structure is merely an example used to describe the capability information of a single camera:

[0154] {

[0155] "cameraId":"rear_camera_01",

[0156] "position":"rear",

[0157] "orientation":"backward",

[0158] "fieldOfView":{"horizontal":150,"vertical":95},

[0159] "resolution":{"width":1920,"height":1080},

[0160] "frameRate":30,

[0161] "lowLightLevel":0.82,

[0162] "dynamicRangeLevel":0.76,

[0163] "stabilityLevel": 0.88,

[0164] "distortionCorrection":true,

[0165] "occlusionRate": 0.05,

[0166] "latencyMs":24,

[0167] "syncState":"locked",

[0168] "available":true,

[0169] "capabilityTags":["parking_area","vehicle_boundary","rear_obstacle"]

[0170] }

[0171] The system calculates the contribution score of each camera to the current subject task based on subject parameters and camera capability descriptions. The contribution score can be obtained by combining the following factors: whether the target subject is located in the effective coverage area of ​​the camera, the proportion of the subject area in the camera image, image clarity, exposure, degree of occlusion, time synchronization feasibility, current latency, whether there is redundant view with other cameras, and whether it is a necessary view for the current subject.

[0172] To make the above factors computable, the system first categorizes them into several intermediate sub-items, and then calculates the comprehensive score. For example, whether the target subject is located in the effective coverage area, the proportion of the subject area, and whether it is a necessary viewing angle for the current theme can be used to form the theme relevance score R; image clarity, exposure, and degree of occlusion can be used to form the image quality score Q; and time synchronization feasibility, current latency, and redundancy or complementarity with other cameras can be used to form the viewpoint synergy complementarity score C.

[0173] In Example 7, in one embodiment, the system uses a weighted scoring framework to rank each camera. The overall score is obtained by weighting the subject relevance score, image quality score, and viewpoint complementarity score. For ease of implementation, the overall score is set as S = α × R + β × Q + γ × C, where R represents the subject relevance score, Q represents the image quality score, and C represents the viewpoint complementarity score. R, Q, and C can be normalized to a range of 0-100 points. Further, R can be calculated from the target subject pixel area ratio A, the subject feature hit rate F, and the event window matching degree T. When A is between 18% and 65%, F is not less than 60%, and the proportion T of candidate frames falling into the center interval of the target event window is not less than 70%, R can be increased to above 80 points, triggering a higher weight of α of 0.45-0.60. Q can be calculated from the effective sharp pixel ratio P, the total overexposed / underexposed pixel ratio E, and the occlusion area ratio O; when P is not less than 75%, E is not more than 12%, and O is not more than 10%, Q can be improved to over 80 points. C can be calculated from the newly added visible area ratio M, the cross-camera timestamp difference Δt, and the redundant area ratio G; when M is not less than 20%, Δt is not more than 80ms, and G is not more than 50%, C can be improved to over 70 points. Conversely, when A is less than 8% or F is less than 35%, the system can limit R to below 40 points and lower α to no more than 0.30 to avoid weakly correlated scenes being mistakenly selected due to high local sharpness.

[0174] In Example 8, preferably, the system does not require the entire preset set of theme features to be hit before triggering the main scene selection. Instead, it sets an effective feature hit interval. For example, for the theme of reversing into a parking space, the parking space line, the target parking space frame, the outline of the rear obstacle, the left gap, the right gap, the vehicle body boundary, the steering wheel correction action, and the low-speed reversing state can be used as the theme feature set. When the candidate sample hits 4-6 of these features, that is, the feature hit rate is between 50% and 75%, and it contains at least one spatial boundary feature and one dynamic behavior feature, it can be considered as the preferred trigger interval. At this time, both theme relevance and key moments in the early or middle stages of the event development can be preserved. If the hit rate exceeds 90%, it usually indicates that the system has waited for too many features to appear simultaneously, the candidate frame is closer to the end of the event or the vehicle is basically stationary, and the system can deduct 10-20 points from the time freshness to prevent missing the best display opportunity. If the hit rate is less than 25%, the sample can be determined to contain only scattered or invalid features, which are likely to deviate from the subject capture target. The system marks it as a low contribution sample or low contribution camera and reconstructs the storyboard template based on the remaining candidate cameras. If a certain viewpoint that must be retained is missing, the system calls the alternative viewpoint rule, that is, to use adjacent direction cameras or cross time frames to compensate for the information required for that viewpoint.

[0175] Example 9, (III) Theme-Perspective Requirement Modeling and Protagonist Switching Mechanism

[0176] This invention further establishes a theme-viewpoint requirement model to determine the main subject and viewpoint composition under different themes. The main subject can be the driver, a specific target outside the vehicle, a road risk area, a vehicle attitude change area, or the state of the occupants inside the vehicle. After identifying the main subject based on the theme parameters, the system assigns a main storyboard to the main subject and uses other information strongly related to the theme as auxiliary storyboards.

[0177] For example, under the theme of "Recording the correlation between driver status and road conditions", the main storyboard can be assigned to the driver monitoring camera, and the auxiliary storyboards can be assigned to the forward-looking road camera and the instrument panel area respectively; under the theme of "Recording the process of avoiding external obstacles", the main storyboard can be assigned to the forward-looking or side-looking camera that best reflects the spatial relationship of obstacles, and the auxiliary storyboard is used to supplement the vehicle boundary and the driver's operation reaction; under the theme of "Summary of parking process", the main storyboard can be assigned to the rear-view or surround-view camera, and the auxiliary storyboard is used to show the left and right gaps and the driver's observation actions.

[0178] Example 10: In one embodiment, the topic-perspective requirement model can be implemented using a mapping table, a rule engine, or a combination of both. The following mapping rule table is merely an example:

[0179] Theme type Main character Must retain perspective A perspective that can be discarded illustrate Reverse parking Parking area or parking space boundary Rearview, Surround View Forward Prioritize ensuring visibility of the relationship between rear obstacles and vehicle boundaries. Fatigue driving warning Driver's facial expression Driver monitoring, forward-looking road Rearview A display that considers both the driver's condition and the road conditions ahead Meeting on a narrow road Oncoming vehicles and lateral clearance Forward vision, adjacent lateral vision Distant view Prioritize reflecting the spatial relationships and risk points of vehicles meeting each other. Forward Collision Warning Ahead risk targets Forward-looking driver monitoring Rearview or risk-independent perspective Emphasis on hazardous targets, driver reactions, and vehicle responses

[0180] Furthermore, the rule engine can adopt the following logical expression: when the topic type is reversing into a parking space and the safety priority is not lower than a preset threshold, the parking area is set as the main object, the rear view and surround view are set as mandatory viewing angles, and the front view is set as a discardable viewing angle; when the topic type is switched to fatigue driving reminder, the driver monitoring and the front view road are set as mandatory viewing angles. Through the above mapping mechanism, the system can quickly output the set of viewing angle requirements and the filtering priority after the topic is determined.

[0181] Example 11, (IV) Storyboard Template Generation and Adaptive Degradation Mechanism

[0182] After determining the camera roles, the storyboard template generation module generates a corresponding storyboard layout template based on the vehicle's screen size, output resolution, number of available cameras, subject priority, and target subject location. The template includes at least a main storyboard area and one or more auxiliary storyboard areas, and the shape, area, order, and boundaries of each area can be dynamically adjusted according to the current task.

[0183] When there are a large number of available cameras, the system uses a complete template that includes a main storyboard, side supplementary storyboards, close-up storyboards, and panoramic storyboards. When the number of available cameras decreases, the system automatically reduces the number of auxiliary storyboards, expands the area of ​​the main storyboard, and merges or removes areas with lower information value. When only two or three high-contribution cameras are retained, the system uses a simplified template to ensure that key subjects are displayed completely. If necessary, the system can also define the order of storyboards according to the theme script to give the result a clear narrative logic.

[0184] For adapting the same theme to different vehicle models, this invention does not require a fixed number of storyboards to be output. Instead, it dynamically determines the template complexity based on the number of available cameras, capability scoring results, and theme requirements. For example, in the reversing parking theme, a vehicle with a more complete configuration can output a main storyboard, two side auxiliary storyboards, a surround view storyboard, and a driver observation storyboard. In vehicles with fewer cameras, the system can automatically reduce the template to a simplified main storyboard and one auxiliary storyboard, and increase the area of ​​the main storyboard, thereby maintaining the usability of the results and the consistency of the theme across different vehicle models.

[0185] Example 12, (V) Synchronous Acquisition Mode

[0186] When the subject requires reflecting the status from multiple angles at the same time, the system enters synchronous acquisition mode. The synchronous acquisition module uses the vehicle's unified clock or unified time base as a reference to time-align the image frames from multiple cameras, and sends the set of frames that meet the synchronization conditions to the scene composition module.

[0187] In Example 13, in one embodiment, the system can set a time deviation threshold, for example, no more than 50 milliseconds. If a certain camera fails to acquire a valid synchronized frame within the allowable deviation range, the system can select the nearest available frame within a limited time window and add a non-strict synchronization mark to that scene. If the deviation exceeds the threshold or there are no available frames, an alternative perspective is enabled or the priority of that scene is reduced in the template. The synchronized acquisition mode is suitable for topics such as instantaneous panoramic recording of lane changes, instantaneous spatial relationship recording of oncoming traffic, and joint display of drivers and roads during emergency braking.

[0188] Furthermore, the system can determine the acquisition mode based on the time window field, instantaneous recording identifier, and event process display requirements in the topic parameters. When the time window length is 0, the time window length is less than or equal to the preset instantaneous threshold, or the topic only requires recording the state at a certain moment, the system enters the synchronous acquisition mode; when the time window length is greater than the preset instantaneous threshold and the topic parameters include event process display requirements, process summary requirements, or script-driven orchestration requirements, the system enters the cross-temporal keyframe joint orchestration mode.

[0189] Example 14, (VI) Cross-temporal keyframe joint orchestration mode

[0190] When the theme requires showcasing the event process rather than a single moment, the system enters a cross-temporal keyframe joint arrangement mode. This mode can be implemented in the following steps: First, an event time window is established based on the triggering event; then, candidate frames within this time window are extracted from the caches of each camera; next, the candidate frames are scored based on theme parameters, script rules, changes in the target subject's position, sharpness, event criticality, and viewpoint complementarity; finally, keyframes from different time points before, during, and after the event are selected and mapped to the main storyboard and auxiliary storyboards respectively, forming a single storyboard image or an ordered storyboard sequence.

[0191] The cross-temporal keyframe joint orchestration mode can be directly triggered by the event process display requirements in the theme parameters, or it can be automatically triggered by the safety event type, process summary theme, or preset script rules. For example, when themes such as current collision warning, narrow road meeting process recording, or reversing parking summary require the retention of continuous stage information before, during, and after the event, the system can automatically switch to the cross-temporal keyframe joint orchestration mode to avoid the loss of process information caused by retaining only a single instant frame.

[0192] For example, in the "Summary of Reversing into a Parking Space" topic, the system can combine the rear-view keyframes when the vehicle approaches the parking space, the side-view keyframes when the vehicle body is close to the boundary, and the surround-view keyframes when the vehicle is finally parked into the space into a single storyboard. In the "Record of a Narrow Road Meeting Event" topic, the system can select and arrange the forward-view keyframes before the oncoming vehicle is detected, the side-view keyframes with the smallest left-right distance during the meeting, and the subsequent keyframes after the vehicle has passed. This mechanism solves the problem that existing solutions can only output simultaneous, same-frame sequences, making it difficult to express the complete event process.

[0193] Example 15, (VII) Image Preprocessing and Storyboard Synthesis Mechanism

[0194] The storyboard compositing module performs at least one preprocessing operation on the selected footage, including distortion correction, target area cropping, brightness adjustment, noise reduction and enhancement, boundary line generation, text label overlay, and theme style rendering. These processes prioritize ensuring the identifiability of the main subject, clarity of event relationships, and stability of the storyboard structure. Theme style rendering can be an optional step, used to generate documentary, summary, or cartoon-style results based on the theme.

[0195] During the compositing process, the system prioritizes displaying the main storyboard with the most relevant subject matter based on the area, position, and hierarchy of each region in the storyboard template. Secondary storyboards supplement spatial relationships, driver status, vehicle operation, or environmental context. When necessary, the system can add time sequence markers, event markers, or perspective source markers to the storyboards to facilitate subsequent understanding and review by the user.

[0196] Example 16, (VIII) Closed-loop mechanism for work quality assessment and training

[0197] To achieve continuous optimization, this invention includes a work quality evaluation module to perform a structured evaluation of each generated storyboard result. Evaluation indicators may include one or more of the following: theme matching score, subject coverage, key event retention rate, perspective redundancy, clarity score, synchronization integrity, user retention rate, user reshoot rate, and user secondary editing frequency (as shown in Table 3).

[0198] Table 3

[0199] Evaluation indicators illustrate Quantification method example Topic matching score The degree of alignment between the results and the theme objectives 0-100 points Main body coverage The visibility ratio of the target subject in the main storyboard and auxiliary storyboards percentage Key event retention rate The proportion of keyframes covering critical event nodes percentage Viewpoint redundancy The degree of information overlap between different scenes Scores from 0 to 100, the lower the better. Clarity rating Clarity of images or video clips 0-100 points Synchronization completeness Multi-channel material time consistency 0-100 points User retention rate The proportion of generated results saved by users percentage User retake rate The proportion of users re-triggered generation percentage User secondary editing frequency Number of times users manually modified the results frequency

[0200] The evaluation results, along with the subject parameters, vehicle status, camera capability data, camera selection results, keyframe selection results, and the final video data, are stored together to form training samples. The shooting control model training module uses these training samples to update the camera scoring parameters, keyframe selection parameters, storyboard template selection parameters, or subject-viewpoint mapping parameters, enabling the generation of better results under similar subjects and vehicle configurations in subsequent iterations.

[0201] This invention enables the analysis of shooting control strategies based on the results of the work, and allows for continuous improvement of system performance with clear optimization directions and evaluation standards. This closed-loop mechanism significantly enhances the system's long-term stability and subject adaptation capabilities.

[0202] Example 17, (ix) Method Flow Overview

[0203] The method flow of this invention can be summarized as follows: Step T1, acquire theme parameters, vehicle status data, and camera capability data; Step T2, construct a theme-viewpoint requirement model to determine the target subject, the required viewpoints, and the viewpoints that can be discarded; Step T3, perform camera scoring, sorting, and role allocation to determine the main storyboard camera, auxiliary storyboard cameras, and low-contribution cameras; Step T4, generate storyboard templates based on the remaining (selected) cameras and theme requirements; Step T5, perform synchronous acquisition or cross-temporal keyframe extraction based on the time window field, instantaneous recording identifier, event process display requirements, or script rules in the theme parameters; Step T6, perform preprocessing and storyboard compositing on the selected image frames to obtain theme-driven storyboard results; Step T7, perform quality evaluation on the final product and feed the data back to the shooting control model to complete the strategy update.

[0204] Through the above process, a complete link is formed between the modules, from topic input, device adaptation, content selection, structure output to feedback optimization, thereby achieving stable, scalable, and reusable vehicle multi-camera storyboard generation capabilities.

[0205] Example 18, Example 1: Recording the correlation between driver status and road conditions ahead

[0206] In a vehicle equipped with a forward-facing camera, a driver monitoring camera, a left-side view camera, a right-side view camera, and a rear-view camera, the system receives a "Fatigue Driving Warning Process Recording" topic. The topic parameters specify the driver as the subject, requiring the retention of the driver's facial expression, the road environment ahead, and at least one lateral supplementary view. After reading the capability descriptions of each camera, the system identifies the driver monitoring camera as having the highest relevance to the topic and assigns it to the main shot; the forward-facing camera is assigned as the first auxiliary shot to display the road environment; the left-side or right-side camera more relevant to the current steering direction is assigned as the second auxiliary shot; and the rear-view camera, due to its lower relevance to the topic, is marked as a low-contribution camera.

[0207] When a driver fatigue detection event is triggered, the system extracts candidate frames within a preset time window before and after the trigger. It then selects keyframes of the driver's face at the moment their eyes are closed or their gaze is deviated, keyframes of the forward-looking road at the corresponding moment, and keyframes of the lateral environment, combining them into a single cross-temporal scene breakdown. This result simultaneously reflects both the driver's state and external road conditions, providing a basis for driving safety recording.

[0208] Example 19, Example 2: Contribution Score Quantification Calculation and Topic-Perspective Mapping Example

[0209] In a vehicle equipped with a front-view camera, a rear-view camera, a left-side view camera, a right-side view camera, and a surround-view camera, the system receives a "Summary of Reversing into Parking Process" topic and sets the parking area as the main object. The topic-view requirement model outputs the following results: the rear-view and surround-view views are mandatory, the left and right side views are optional, the front-view view is optional, and the safety priority is high.

[0210] In this embodiment, the system normalizes the topic relevance score R, image quality score Q, and viewpoint synergy complementarity score C to a range of 0-100, and calculates the contribution score of each camera using a comprehensive score S=α×R+β×Q+γ×C. For the topic of "summary of reversing into a parking space process," the system pre-selects eight types of topic features: parking space lines, target parking space frame, rear obstacle outline, left side gap, right side gap, vehicle body boundary, steering wheel correction action, and low-speed reversing state. Preferably, when the candidate image hits 4-6 types of topic features, the main pixel area ratio is between 20% and 55%, and the deviation between the timestamp and the key event window for parking into the space does not exceed 0.6s, the system can select α=0.50-0.55, β=0.25-0.30, and γ=0.20-0.25; in this embodiment, α=0.50, β=0.30, and γ=0.20 are used. Calculations show that the rearview camera hit five types of features: parking space lines, target parking space frames, rear obstacle outlines, vehicle boundaries, and low-speed reversing conditions, achieving a feature hit rate of 62.5%. The main pixel area ratio was 28%, the effective clear pixel ratio was 82%, the occlusion area ratio was 6%, and the newly added visible area ratio was 22%, resulting in R=92, Q=86, C=74, and corresponding S=86.6. The surround-view camera hit six types of features, achieving a feature hit rate of 75%. The main pixel area ratio was 34%, the effective clear pixel ratio was 79%, the occlusion area ratio was 4%, and the newly added visible area ratio was 31%, resulting in R=88, Q=80, C=86, Q=86, C ... =94, corresponding to S=86.8; the left-side camera hit 4 types of features with a feature hit rate of 50%, and the main pixel area ratio is 16%, resulting in R=76, Q=79, C=70, corresponding to S=75.7; the right-side camera hit 4 types of features with a feature hit rate of 50%, and the main pixel area ratio is 15%, resulting in R=74, Q=81, C=68, corresponding to S=74.9; the front-side camera only hit 2 types of features: steering wheel correction precursor and low-speed reversing state, with a feature hit rate of 25%, and the main pixel area ratio is less than 5%, resulting in R=18, Q=88, C=22, corresponding to S=39.8.

[0211] Based on this, the system identified the surround-view and rear-view cameras as the main storyboard and panoramic supplementary storyboard, respectively, designated the left and right side-view cameras as auxiliary storyboards, and marked the front-view camera as a low-contribution camera. Among them, the rear-view camera most directly expresses the spatial relationship between parking lines and obstacles behind, making it suitable as the main storyboard; although the theme relevance of the surround-view camera is slightly lower than that of the rear-view camera, its newly added visible area ratio reaches 31%, making it more suitable as a panoramic supplementary storyboard; the feature hit rate of the left and right side-view cameras both reach 50%, which can supplement lateral gap information; the front-view camera was excluded from the main storyboard candidate set because its theme feature hit rate is only 25% and the main pixel area accounts for less than 5%.

[0212] Example 20, Example 3: Comparison of Adaptive Storyboards Across Vehicle Models on the Same Theme

[0213] Under the same theme of "Summary of Reversing into Parking Process," the system was applied to two different vehicle configurations. The first configuration included a rearview camera, a surround-view camera, a left-side camera, a right-side camera, and a driver monitoring camera; the second configuration only included a rearview camera, a left-side camera, and a right-side camera. The theme parameters were consistent for both models, emphasizing the parking area, vehicle boundaries, and key corrective actions during the parking process.

[0214] For the first vehicle type, the system determines the main view from the rear-view camera, the two auxiliary view cameras from the left and right sides, the panoramic view camera as the panoramic supplementary view, and the driver monitoring camera as the observation action view, based on the camera capability description and contribution score. This results in a complete template output including the main view, the two auxiliary view cameras, the panoramic view, and the driver observation view. This output simultaneously preserves the relationship to rear obstacles, left and right side clearances, the overall vehicle attitude, and the driver's observation actions.

[0215] For the second vehicle type, after detecting the lack of surround-view cameras and driver monitoring cameras, the system no longer uses the aforementioned complete template. Instead, it recalculates the role allocation based on the remaining available cameras. The system designates the rear-view camera as the main storyboard and selects the path more relevant to the current correction direction or the current minimum clearance from the left-side and right-side cameras as the auxiliary storyboard. Simultaneously, it automatically expands the area of ​​the main storyboard and sets the other side-side camera as a backup viewpoint or rotates it into the auxiliary storyboard at different times. The final output is a simplified template consisting of one main storyboard and one auxiliary storyboard.

[0216] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A subject-driven, vehicle-mounted multi-camera adaptive split-scene shooting method, characterized in that, include: S1. Obtain the topic parameters, vehicle status data, and camera capability data of each vehicle camera for the current task; S2. Construct a topic-perspective requirement model based on the topic parameters to determine the target subject, the necessary perspectives to be retained, and the perspectives that can be discarded for the current task. S3. Based on the camera capability data of each vehicle camera and the topic-viewpoint demand model, perform contribution scoring, priority ranking and role assignment for each vehicle camera, filter out high-contribution cameras and remove low-contribution cameras to determine the usable cameras. S4. Based on the number of available cameras after filtering, the role allocation results and the current topic parameters, dynamically generate an adapted storyboard template, the storyboard template including: storyboard areas with different primary and secondary roles; S5. Based on the task requirements corresponding to the topic parameters, selectively execute the synchronous acquisition mode or the cross-temporal keyframe joint arrangement mode to obtain the frame set of the corresponding valid image. S6. Preprocess and synthesize the frame set of the effective images, and output the vehicle-mounted storyboard result with theme matching. S7. Evaluate the quality of the output storyboard results, and iteratively optimize the shooting control strategy based on the evaluation data.

2. The subject-driven adaptive multi-camera split-scene shooting method for vehicles according to claim 1, characterized in that, The topic parameters include: The theme type, target subject type, target event type, time window, maximum number of storyboards, output style, required viewpoints, optional viewpoints, security priority, and triggering method must be included in at least one of the following: The topic parameters can be manually entered or automatically generated; when automatically generated, they are generated based on the vehicle status. The camera capability data includes: physical attributes, performance indicators, status information, and functional attributes; The physical attributes include: installation location, orientation, field of view, resolution, and frame rate; The performance indicators include: low-light capability, dynamic range, and imaging stability; The status information includes: online status, fault status, obstruction status, and synchronization status; The functional attributes include labels for object detection, face recognition, and behavior recognition capabilities.

3. The subject-driven adaptive multi-camera split-scene shooting method for vehicles according to claim 1, characterized in that, In step S3, the contribution score is calculated as follows: Based on the topic relevance score R, the image quality score Q, and the perspective synergy and complementarity score C, the weighted comprehensive contribution score S is calculated as follows: S = α × R + β × Q + γ × C. Where α, β, and γ are preset weighting coefficients, and R, Q, and C are all normalized to the 0-100 score range; The topic relevance score R is calculated from the target subject pixel area ratio, topic feature hit rate, and event window matching degree. The image quality score Q is calculated from the effective clear pixel ratio, the exposed pixel ratio, and the occlusion area ratio. The viewpoint coordination complementarity score C is calculated from the newly added visible area ratio, cross-camera timestamp difference, and viewpoint redundancy area ratio. It also includes a dynamic weight adjustment mechanism: when the target subject pixel area ratio, topic feature hit rate, and event window matching degree meet the preset high-quality conditions, the weight α corresponding to the topic relevance score is increased. When the hit rate of the subject feature is lower than the preset hit threshold, or / and the effective proportion of the target subject is lower than the preset proportion threshold, the weight α is reduced and the corresponding camera is determined to be a low contribution camera in this round.

4. The subject-driven adaptive multi-camera split-scene shooting method for vehicles according to claim 1, characterized in that, The topic-perspective requirement model includes rules for establishing mappings between topics, main characters, and perspective sets: The main subject object is dynamically switched according to different theme types. The main subject objects include: driver, external target, road risk area, parking area and vehicle posture. The matching process ensures that the viewpoint is retained or discarded, enabling adaptive viewpoint adaptation for different scenarios.

5. The subject-driven adaptive multi-camera split-scene shooting method for vehicles according to claim 1, characterized in that, In step S4, the dynamic generation of the adapted storyboard template includes an adaptive degradation mechanism: When the vehicle-mounted camera configuration is complete and the number of available cameras is sufficient, a complete template including the main storyboard, auxiliary storyboard, supplementary storyboard, and panoramic storyboard is generated. When there are missing, faulty, obstructed, or insufficient cameras, the system automatically reduces the number of auxiliary shots, expands the display area of ​​the main shot, removes redundant shots, and generates a simplified adaptation template. For different vehicle models with heterogeneous camera configurations, the complexity of the storyboard template is dynamically adjusted to maintain the consistency and integrity of the theme display.

6. The subject-driven adaptive multi-camera split-scene shooting method for vehicles according to claim 1, characterized in that, The execution logic of the synchronous acquisition mode includes: Based on the vehicle's unified time, time alignment is performed on image frames from multiple cameras; Among them, a preset time deviation threshold is set to filter and obtain a set of image frames that meet the synchronization conditions at the same time. It also includes: replacing camera frames that exceed the time deviation threshold or fail with the nearest available frame or enabling backup view compensation, and marking non-strict synchronization states. The synchronous acquisition mode is primarily applicable to scenarios involving instantaneous state recording.

7. The subject-driven adaptive multi-camera split-scene shooting method for vehicles according to claim 1, characterized in that, The execution logic of the cross-temporal keyframe joint orchestration mode includes: Determine a preset time window based on the target event, and extract candidate image frames from each camera within the time window; Candidate image frames are scored and filtered based on topic matching degree, image quality, event criticality, and viewpoint complementarity, and key frames before, during, and after the event are selected respectively. Keyframes at different time intervals are mapped to corresponding areas of the storyboard template to form a storyboard sequence that can reconstruct the complete event process.

8. The subject-driven adaptive multi-camera split-scene shooting method for vehicles according to claim 1, characterized in that, The preprocessing described in step S6 includes: At least one of the following: distortion correction, target region cropping, brightness adjustment, noise reduction and enhancement, boundary generation, identifier overlay, and style rendering; The style rendering includes differentiated visual outputs such as documentary style, abstract style, and comic style, and the style rendering process maintains the clarity of the target subject at the preset resolution.

9. The subject-driven adaptive multi-camera split-scene shooting method for vehicles according to claim 1, characterized in that, The quality assessment described in step S7 uses multi-dimensional quantitative indicators: The quantitative indicators include: topic matching score, subject coverage, key event retention rate, perspective redundancy, clarity score, synchronization integrity, and user operation feedback indicators. The evaluation data, subject parameters, camera data, and frame selection data of each storyboard result are associated and stored as training samples; Based on the training samples, the camera scoring parameters, keyframe extraction parameters, and storyboard template parameters are iteratively updated to achieve continuous optimization of the shooting control strategy.

10. A subject-driven, vehicle-mounted multi-camera adaptive split-scene shooting system, characterized in that, For executing the subject-driven vehicle-mounted multi-camera adaptive split-scene shooting method according to any one of claims 1-9, the subject-driven vehicle-mounted multi-camera adaptive split-scene shooting system comprises: The topic parameter parsing module is used to receive or automatically generate topic parameters for the current task; The vehicle status acquisition module is used to acquire vehicle status, environmental and driving event data; The camera capability modeling module is used to build and store the structured capability data of each vehicle camera; The topic-perspective requirements modeling module is used to determine the target subject and perspective requirements set based on topic parameters; The camera scoring and low-contribution screening module is used to score, sort, and remove low-contribution devices from cameras. The storyboard template generation module is used to dynamically generate and adaptively downgrade and adjust the storyboard layout template. The synchronous acquisition module and the time-series keyframe extraction module are used to execute the synchronous acquisition mode and the cross-time-series keyframe orchestration mode, respectively. The storyboard compositing module is used to complete image preprocessing and output the compositing results of the storyboard. The work quality assessment module and the shooting control model training module are used to complete the closed loop of result quantitative evaluation and strategy iterative optimization; Each of the modules is deployed in the vehicle central control terminal, the cockpit computing platform, or the vehicle domain controller.

Images