A fall detection method, apparatus and device based on video data

Abstract

The application relates to the field of fall detection, and discloses a fall detection method, device and equipment based on video data, the method comprising the following steps: acquiring video data and performing preprocessing; adopting a preset detection model to perform human target detection and tracking on the video data, constructing a human continuous behavior sequence of at least one target object, and extracting behavior semantic features of each action; calling a preset judgment rule and / or a preset classification model and generating a structured fall judgment result, a fall early warning instruction and / or a fall intervention scheme based on the behavior semantic features. Through a knowledge transfer and a semi-supervised labeling mechanism, the application reduces the dependence on artificial labeling data, enhances the adaptability and stability to complex scenes; meanwhile, through joint analysis of human posture changes, motion time sequence features and subject features, high-precision identification and timely early warning of a fall event are realized, the detection accuracy and usability are improved, and stable and reliable operation in various application scenarios is achieved.

Description

Technical Field

[0001] This invention relates to the field of fall detection, and more particularly to a fall detection method, apparatus, and device based on video data. Background Technology

[0002] As the population ages, the safety of the elderly and those with limited mobility in their homes, nursing homes, and medical facilities is receiving increasing attention. Falls, in particular, are characterized by their sudden occurrence, high risk of injury, and frequent delays in detection, severely impacting the safety and quality of life of those under supervision. Current fall monitoring technologies primarily include manual inspections, wearable monitoring devices, and video-based automated detection methods. Manual inspections are costly and cannot guarantee 24 / 7 monitoring; wearable devices rely on the wearer's active cooperation, raising concerns about compliance and device maintenance; video-based methods have a natural advantage in reducing reliance on wearable devices, but often require large amounts of labeled data to ensure model performance in different scenarios.

[0003] Currently, video-based fall detection technology still faces several challenges in practical applications: First, it is sensitive to specific monitoring scenarios, such as changes in lighting, occlusion, camera angle, and background complexity, which leads to decreased detection stability; second, its reliance on a large number of manually labeled samples makes rapid deployment in new scenarios difficult and costly; and third, it has limited functionality in terms of interpretability of detection results and generation of follow-up handling suggestions, making it difficult to directly provide sufficient support for rescue decisions. Summary of the Invention

[0004] This invention provides a fall detection method, apparatus, and device based on video data, aiming to achieve high-precision identification, interpretable analysis, and early warning output of fall events through joint analysis of human posture changes and movement temporal characteristics, thereby solving one of the technical problems in related technologies.

[0005] A first aspect of this invention provides a fall detection method based on video data, comprising the following steps: Step 1: Acquire video data of the target scene and perform preprocessing; Step 2: Use a preset detection model to detect and track human targets in the video data, construct a continuous human behavior sequence of at least one target object, and extract the behavioral semantic features of each action in the continuous human behavior sequence. The behavioral semantic features include human posture features and motion temporal features. Step 3: Invoke the preset judgment rules and / or preset classification model and make a fall judgment based on the behavioral semantic features, and generate a structured fall judgment result, which includes at least one of the fall risk level, fall type and fall posture; Step 4: Generate a fall warning instruction and / or a fall intervention plan based on the fall judgment result.

[0006] A second aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the fall detection method based on video data described above.

[0007] A third aspect of the present invention provides a fall detection device based on video data, including a computer-readable storage medium and a processor, wherein the processor executes a computer program on the computer-readable storage medium to implement the steps of the fall detection method based on video data described above.

[0008] A fourth aspect of this invention provides a fall detection device based on video data, including a data acquisition module, a behavior state analysis module, a fall detection module, and a result output module. The data acquisition module is used to acquire and preprocess video data of the target scene; The behavior state analysis module is used to perform human target detection and tracking on the preprocessed video data using a preset detection model, construct at least one continuous human behavior sequence of the target object, and extract the behavioral semantic features of each action in the continuous human behavior sequence. The behavioral semantic features include human posture features and motion temporal features. The fall detection module is used to call preset judgment rules and / or preset classification models and make fall judgments based on the behavioral semantic features, and generate structured fall judgment results, which include at least one of fall risk level, fall type and fall posture; The result output module is used to generate fall warning instructions and / or fall intervention plans based on the fall judgment results.

[0009] The beneficial effects of this invention are as follows: This invention provides a fall detection method, apparatus, and device based on video data. Through knowledge transfer and semi-supervised annotation mechanisms, it significantly reduces the dependence on manually labeled data and the cost of manual annotation, and enhances the adaptability and stability of the algorithm to complex scenarios. At the same time, through joint analysis of human posture changes and motion temporal features, it achieves high-precision identification and timely early warning output of fall events, improving detection accuracy and practical usability, so as to operate stably and reliably in various application scenarios such as smart elderly care, medical monitoring, and public safety.

[0010] To make the above-mentioned objects, features and advantages of the invention more apparent and understandable, preferred embodiments of the invention are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating a fall detection method based on video data provided in one embodiment; Figure 2 This is a flowchart illustrating a fall detection method based on video data provided in another embodiment; Figure 3 This is a schematic diagram of a fall detection result provided in one embodiment; Figure 4 This is a schematic diagram of a fall detection device based on video data provided in one embodiment; Figure 5 This is a schematic diagram of the structure of a fall detection device based on video data provided in one embodiment. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0014] It should be noted that, unless otherwise specified, the various features in the embodiments of this invention can be combined with each other, all of which are within the protection scope of this invention. Furthermore, although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than the module division in the device or the order in the flowchart. Moreover, the terms "first," "second," and "third" used in this invention do not limit the data or execution order, but only distinguish identical or similar items with essentially the same function and effect.

[0015] Figure 1 This is a flowchart illustrating a fall detection method based on video data provided in Embodiment 1. Figure 1 As shown, it includes the following steps: Step 1: Acquire video data of the target scene and perform preprocessing.

[0016] For example, video data of the target scene can be collected by monitoring equipment such as cameras, and the video data can be initially encoded (such as H.264 encoding), denoised and standardized, and then video frame parsing can be performed to form a video frame sequence arranged in chronological order, so as to facilitate subsequent human region detection and tracking.

[0017] Step 2: Use a preset detection model to detect and track human targets in the video data, construct a continuous human behavior sequence of at least one target object, and extract the behavioral semantic features of each action in the continuous human behavior sequence. The behavioral semantic features include human posture features and motion temporal features.

[0018] For example, a continuous sequence of human behavior consists of basic actions such as standing, walking, sitting, and lying down. Human posture features include at least the orientation changes, posture angle changes, and relative height changes of several key human body points. Motion temporal features include at least the instantaneous velocity, acceleration, duration of action, and direction of motion for each action. These features can all be obtained by performing human region detection and tracking and key point pose estimation on a continuous sequence of video frames.

[0019] Step 3: Invoke preset judgment rules and / or preset classification models and perform fall judgment based on the behavioral semantic features to generate a structured fall judgment result. The fall judgment result includes at least one of fall risk level, fall type, and fall posture. For example, when a significant postural abnormality occurs within a short period of time and is accompanied by features such as instantaneous speed exceeding a preset threshold, a fall event can be determined.

[0020] Step 4: Generate fall warning instructions and / or fall intervention plans based on the fall judgment results to achieve automatic identification, warning and recording of fall events.

[0021] The above embodiments provide a fall detection method based on video data. Based on human posture changes and movement timing features, it can achieve high-precision identification of fall events and timely early warning output, improving detection accuracy and practical usability, so as to operate stably and reliably in various application scenarios such as smart elderly care, medical monitoring and public safety.

[0022] The following combination Figure 2 , Figure 3 The above methods will be explained in detail.

[0023] Figure 2 The overall processing procedure of the fall detection method described in this invention is illustrated. First, continuous video data is received from a video acquisition device, and the continuous video data is parsed and processed to represent the input video as a sequence of video frames arranged in chronological order: , in, This represents a video image frame captured at time t.

[0024] When processing video frame sequences, this invention does not analyze individual frames in isolation, but rather models human behavior as a whole within the video to ensure a continuous depiction of the evolution of human movements. To this end, this invention introduces a video understanding mechanism based on a pre-defined multimodal large model, using a pre-defined structured cue word chain. The continuous actions of at least one target object in a video frame sequence are analyzed. By matching preset atomic actions, the continuous video is decomposed into a target action chain corresponding to each target object. This target action chain consists of multiple atomic actions, as follows:

[0025] in, This represents the atomic action at time t.

[0026] In this way, the system can represent complex human behavior processes as a series of atomic actions with clear semantics, providing a clear temporal structure for subsequent fall risk analysis.

[0027] Then, for each atomic action in the action chain... Based on biomechanics and pre-set behavioral risk warning templates Human posture features and motion timing features of each action in the target action chain are extracted to perform a risk assessment of the action, and a risk feature vector is constructed as follows: , In the formula, Indicates the predicted time of a potential collision. Indicates the degree of abnormality in human posture. Indicates body parts that may come into contact or collide. This indicates the instantaneous speed of the human body during that action. These characteristics, which indicate the potential impact intensity, together constitute a multi-dimensional characterization of the safety risks associated with human movement.

[0028] Then, based on the aforementioned risk feature vector, the conditional probability of falling for each action can be output as follows: , The above steps guide the model to focus on human posture, movement state, and possible collision scenarios, thereby achieving a precise assessment of the risk of falling during a single action.

[0029] Figure 3This illustrates the logical relationship between posture analysis and fall detection. This embodiment uses the target action chain as the main analysis line, jointly modeling human posture changes and motion temporal characteristics. By accumulating and analyzing the risk information of multiple atomic actions in the target action chain, it achieves a judgment of the overall behavioral state.

[0030] Specifically, in the time dimension, the fall conditional probabilities corresponding to each atomic action in the target action chain A are weighted and aggregated, and the comprehensive judgment function is expressed as:

[0031] in, This represents the weight of the t-th atomic action in the overall behavior, used to reflect the degree of influence of different actions on the fall outcome. To preset the fall detection threshold, This is an indicator function.

[0032] Finally, the fall risk level corresponding to the target object is generated based on the numerical range of the weighted aggregation calculation result. For example, in one embodiment, to improve the ability to express the degree of human behavioral risk, the final fall determination result is defined as a three-level output, with the following correspondence:

[0033] When the output result is 0, it indicates that the human body is in a normal activity state; when the output result is 0.5, it indicates that abnormal posture or movement characteristics have been detected but the fall judgment conditions have not yet been met; when the output result is 1, it indicates that the system has confirmed that a fall event has occurred, and then sends the fall judgment result, including the fall location, fall risk level, fall type and / or fall posture, as well as subsequent first aid measures and environmental optimization suggestions, to the emergency contact via SMS or email according to the user settings, and confirms whether the notification has been delivered.

[0034] The above embodiments, by introducing the multi-level behavioral state determination mechanism, can accurately identify fallen events that have already occurred, and can also mark the identified high-risk behaviors in advance before a fall occurs, thereby providing a time window for early warning and intervention.

[0035] For example, in a preferred embodiment, the conditional probability of falling corresponding to each action is generated based on the risk feature vector, specifically as follows: Obtain the main and accompanying judgment features of the risk feature vector and their corresponding fall judgment thresholds. For example, set the degree of abnormal human posture as the main judgment feature and the instantaneous movement speed or impact intensity as the accompanying judgment feature.

[0036] The system invokes a preset judgment rule and generates the fall conditional probability by comprehensively considering the comparison results between the main judgment feature and / or the accompanying judgment feature and the preset judgment rule. The preset judgment rule adopts threshold-based judgment, classification model-based judgment, or a fusion of both, thereby further improving the accuracy of fall detection results.

[0037] Exemplary, in a preferred embodiment, such as Figure 2 As shown, in determining whether a fall has occurred, in addition to considering human posture characteristics and movement sequence characteristics, the subject's characteristics are also considered, specifically including the following steps: S301, at least one subject feature of a target object is obtained through the video data. The subject feature includes at least one of age, gender, and assistive tool. Specifically, the subject feature can be obtained by performing face recognition, assistive tool detection, etc., on video frame images, or by matching face recognition with preset user information to obtain the subject feature.

[0038] S302, optimize the identification results of the risk feature vector and / or the fall conditional probability based on the subject features. For example, when the subject features meet the preset conditions, adjust the fall judgment threshold, or mark the subject features of the training set during the training process of the classification model, so as to combine human posture features, motion sequence features and subject features for comprehensive judgment and improve the reliability of fall judgment.

[0039] In an exemplary preferred embodiment, a fall prediction step is further included, specifically: S401, Obtain the corresponding high-risk threshold based on the main characteristics, and generate high-risk judgment conditions; S402, when any action or multiple consecutive actions in the target action chain meet the high-risk judgment condition, the action is marked and a fall warning is issued in advance; The high-risk judgment conditions include the probability of falling in a series of consecutive actions increasing sequentially and / or the probability of falling in at least one action being greater than the high-risk threshold, thereby providing timely warnings and taking measures before a user falls, further ensuring user safety.

[0040] For example, a preferred embodiment introduces a knowledge transfer modeling method during the model building process, extracts behavioral semantic feature distribution and fall judgment information from existing scenarios or historical data, and maps or fine-tunes them to the target scenario to build a training set for training the preset detection model and / or the preset classification model, thereby reducing the target scenario's dependence on a large number of manually labeled samples and improving the generalization ability in new scenarios.

[0041] For example, when the number of manually labeled samples is limited, a preferred embodiment can also adopt a semi-supervised data iterative labeling method. That is, the current model is used to predict falls in unlabeled video data, and the fall prediction results that meet the confidence conditions are selected as pseudo-labels. After manual review and / or confidence threshold screening, they are gradually added to the training set, and the model parameters are iteratively updated, thereby improving the model performance and expanding the effective training samples, significantly reducing the cost of manual labeling, and enhancing the adaptability and stability of the algorithm to complex scenes.

[0042] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0043] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the fall detection method based on video data described above.

[0044] Figure 4 This is a schematic diagram of the fall detection device based on video data provided in Embodiment 2, as shown below. Figure 4 As shown, it includes a data acquisition module 100, a behavior state analysis module 200, a fall detection module 300, and a result output module 400. The data acquisition module 100 is used to acquire and preprocess video data of the target scene; The behavior state analysis module 200 is used to perform human target detection and tracking on the preprocessed video data using a preset detection model, construct at least one continuous human behavior sequence of the target object, and extract the behavioral semantic features of each action in the continuous human behavior sequence. The behavioral semantic features include human posture features and motion temporal features. The fall detection module 300 is used to call preset judgment rules and / or preset classification models and make fall judgments based on the behavioral semantic features, and generate structured fall judgment results, the fall judgment results including at least one of fall risk level, fall type and fall posture; The result output module 400 is used to generate fall warning instructions and / or fall intervention plans based on the fall judgment results.

[0045] The above embodiments provide a fall detection device based on video data. The various modules of the fall detection device are connected through standardized interfaces to collaboratively implement the above-mentioned method steps. Based on the changes in human posture and the timing characteristics of movement, it can achieve high-precision identification of fall events and timely early warning output. This not only improves the detection accuracy and practical usability, but also adopts a modular system design, making the deployment process simple and quick, so as to operate stably and reliably in various application scenarios such as smart elderly care, medical monitoring and public safety.

[0046] In a preferred embodiment, the behavior state analysis module 200 specifically includes: The parsing unit is used to call a preset multimodal large model, and parse the continuous actions of at least one target object in the video frame sequence based on a preset structured cue word chain, and construct the target action chain corresponding to each target object by matching preset atomic actions; The feature extraction unit is used to extract the human posture features and motion sequence features of each action in the target action chain based on the preset behavior risk warning template, and to construct a risk feature vector.

[0047] In a preferred embodiment, the fall detection module 300 specifically includes: The calculation unit is used to generate the fall conditional probability corresponding to each action based on the risk feature vector, and to perform weighted aggregation calculation on the fall conditional probabilities of all actions in the target action chain. The rating unit is used to generate the fall risk level corresponding to the target object based on the numerical range of the weighted aggregation calculation result.

[0048] In a preferred embodiment, the fall detection device further includes a fall prediction module, specifically comprising: The condition generation unit is used to obtain the corresponding high-risk threshold based on the subject characteristics and generate high-risk judgment conditions. The condition judgment unit is used to mark the action and issue an early fall warning when any action or multiple consecutive actions in the target action chain meet the high-risk judgment condition. The high-risk judgment criteria include the conditional probability of falling in a series of consecutive actions increasing sequentially and / or the conditional probability of falling in at least one action being greater than the high-risk threshold.

[0049] It should be noted that the foregoing explanation of the fall detection method embodiment based on video data also applies to the fall detection device based on video data in the above embodiments, and will not be repeated here.

[0050] For example, a preferred embodiment may further include an edge computing module. In the specific deployment process, the data acquisition module 100 and the edge computing module can be deployed on the client side, while the behavior state analysis module 200, the fall detection module 300, and the result output module 400 can be deployed on a cloud server. Specifically, after the data acquisition module 100 collects video data of human activity in the target scene, it encodes and reduces noise through the edge computing module, while simultaneously performing video frame parsing, basic feature extraction, and intermediate result caching, thereby reducing network transmission pressure and improving system real-time performance. When the edge computing module detects suspected abnormal behavior or requires higher-precision analysis, it uploads the relevant video segments and intermediate feature data to the cloud server to perform computationally intensive tasks such as action chain parsing, risk assessment, and fall determination, thereby obtaining more accurate and diverse structured fall detection results (such as structured JSON results) and encapsulating them. The specific content of the fall detection results may include event occurrence time, behavior state level, risk characteristic information, and event records of analysis conclusions, etc. The event logs can be pushed to the upper-layer business system via interface calls, message queues, or communication platforms, and trigger corresponding alarm mechanisms according to the configuration to prompt relevant personnel to conduct on-site verification or handling in a timely manner, so as to adapt to the response process of different deployment scenarios.

[0051] Through the above implementation methods, the present invention realizes a fall detection process based on video data. By using action chain analysis and multi-dimensional risk aggregation mechanism, it reduces the dependence on manual annotation while improving the interpretability of detection results and the applicability of the system.

[0052] This invention also provides a fall detection device based on video data, including a computer-readable storage medium and a processor. When the processor executes a computer program on the computer-readable storage medium, it implements the steps of the fall detection method based on video data described above.

[0053] Figure 5 This is a schematic diagram of the fall detection device based on video data provided in Embodiment 3 of the present invention, as shown below. Figure 5 As shown, the fall detection device 8 based on video data in this embodiment includes: a processor 80, a readable storage medium 81, and a computer program 82 stored in the readable storage medium 81 and executable on the processor 80. When the processor 80 executes the computer program 82, it implements the steps in the various method embodiments described above, for example... Figure 1 The steps shown. Alternatively, when the processor 80 executes the computer program 82, it implements the functions of each module in the above-described device embodiments, for example... Figure 4 The functions of the module shown.

[0054] For example, the computer program 82 may be divided into one or more modules, which are stored in the readable storage medium 81 and executed by the processor 80 to perform the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 82 in the video data-based fall detection device 8.

[0055] The video data-based fall detection device 8 may include, but is not limited to, a processor 80 and a readable storage medium 81. Those skilled in the art will understand that... Figure 5 This is merely an example of a fall detection device 8 based on video data and does not constitute a limitation on the fall detection device 8 based on video data. It may include more or fewer components than shown, or combine certain components, or different components. For example, the fall detection device based on video data may also include a power management module, a computing module, input / output devices, network access devices, buses, etc.

[0056] The processor 80 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0057] The readable storage medium 81 can be an internal storage unit of the video data-based fall detection device 8, such as a hard drive or memory of the video data-based fall detection device 8. The readable storage medium 81 can also be an external storage device of the video data-based fall detection device 8, such as a plug-in hard drive, SmartMediaCard (SMC), SecureDigital (SD) card, or FlashCard equipped on the video data-based fall detection device 8. Furthermore, the readable storage medium 81 can include both internal storage units and external storage devices of the video data-based fall detection device 8. The readable storage medium 81 is used to store the computer program and other programs and data required by the video data-based fall detection device. The readable storage medium 81 can also be used to temporarily store data that has been output or will be output.

[0058] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0059] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0060] Those skilled in the art will recognize that the units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0061] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0062] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0063] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0064] The present invention is not limited to the description in the specification and embodiments, and thus other advantages and modifications can be readily realized by those skilled in the art. Therefore, the present invention is not limited to the specific details, representative devices and illustrated examples shown and described herein without departing from the spirit and scope of the general concept as defined by the claims and their equivalents.

Claims

1. A fall detection method based on video data, characterized in that, Includes the following steps: Step 1: Acquire video data of the target scene and perform preprocessing; Step 2: Use a preset detection model to detect and track human targets in the video data, construct a continuous human behavior sequence of at least one target object, and extract the behavioral semantic features of each action in the continuous human behavior sequence. The behavioral semantic features include human posture features and motion temporal features. Step 3: Invoke the preset judgment rules and / or preset classification model and make fall judgments on single actions and continuous behaviors based on the behavioral semantic features to generate structured fall judgment results. The fall judgment results include at least one of fall risk level, fall type and fall posture. Step 4: Generate a fall warning instruction and / or a fall intervention plan based on the fall judgment result.

2. The fall detection method based on video data according to claim 1, characterized in that, The human posture features include at least several key points on the human body, including changes in orientation, posture angle, and relative height; the motion timing features include at least the corresponding instantaneous velocity, acceleration, duration of motion, and direction of motion.

3. The fall detection method based on video data according to claim 1, characterized in that, In the process of model building, a knowledge transfer modeling method is introduced to extract the distribution of behavioral semantic features and fall judgment information from existing scenarios or historical data, and map or fine-tune them to the target scenario to build a training set for training the preset detection model and / or the preset classification model. And / or use the current model to predict falls in unlabeled video data, and select fall prediction results that meet the confidence conditions as pseudo-labels. After manual review and / or confidence threshold screening, they are gradually added to the training set, and the model parameters are iteratively updated to improve model performance and expand the effective training samples.

4. The fall detection method based on video data according to any one of claims 1-3, characterized in that, The fall risk level is generated as follows: A preset multimodal large model is invoked, and the continuous actions of at least one target object in the video frame sequence are parsed based on a preset structured cue word chain. The target action chain corresponding to each target object is constructed by matching preset atomic actions. Based on the preset behavioral risk warning template, extract the human posture features and motion sequence features of each action in the target action chain, and construct a risk feature vector; The fall conditional probability is generated for each action based on the risk feature vector, and a weighted aggregation calculation is performed on the fall conditional probabilities of all actions in the target action chain. The fall risk level corresponding to the target object is generated based on the numerical range of the weighted aggregation calculation result.

5. The fall detection method based on video data according to claim 4, characterized in that, Risk feature vector at time t Defined as: ， in, Indicates the predicted time of a potential collision. Indicates the degree of abnormality in human posture. Indicates body parts that may come into contact or collide. This indicates the instantaneous speed of the human body during that action. This indicates the potential impact strength.

6. The fall detection method based on video data according to claim 5, characterized in that, Based on the risk feature vector, the conditional probability of falling corresponding to each action is generated as follows: Obtain the main and accompanying judgment features of the risk feature vector, as well as the corresponding fall judgment thresholds; The preset judgment rule is invoked, and the fall conditional probability is generated by comprehensively considering the comparison results between the main judgment feature and / or the accompanying judgment feature and the preset judgment rule. The preset judgment rule adopts threshold-based judgment, classification model-based judgment, or a fusion of both.

7. The fall detection method based on video data according to claim 4, characterized in that, The generation of the fall risk level also includes: At least one subject feature of a target object is obtained from the video data, and the subject feature includes at least one of age, gender, and assistive tool; The identification results of the risk feature vector and / or the fall conditional probability are optimized based on the subject characteristics.

8. The fall detection method based on video data according to claim 7, characterized in that, It also includes a fall prediction step, specifically: Based on the subject characteristics, obtain the corresponding high-risk threshold and generate high-risk judgment conditions; When any action or multiple consecutive actions in the target action chain meet the high-risk judgment condition, the action is marked and a fall warning is issued in advance; The high-risk judgment criteria include the conditional probability of falling in a series of consecutive actions increasing sequentially and / or the conditional probability of falling in at least one action being greater than the high-risk threshold.

9. A fall detection device based on video data, based on the method of any one of claims 1-8, characterized in that, It includes a data acquisition module, a behavior state analysis module, a fall detection module, and a result output module. The data acquisition module is used to acquire and preprocess video data of the target scene; The behavior state analysis module is used to perform human target detection and tracking on the video data using a preset detection model, construct at least one continuous human behavior sequence of the target object, and extract the behavioral semantic features of each action in the continuous human behavior sequence. The behavioral semantic features include human posture features and motion temporal features. The fall detection module is used to call preset judgment rules and / or preset classification models and make fall judgments based on the behavioral semantic features, and generate structured fall judgment results, which include at least one of fall risk level, fall type and fall posture; The result output module is used to generate fall warning instructions and / or fall intervention plans based on the fall judgment results.

10. A fall detection device based on video data, comprising a computer-readable storage medium and a processor, characterized in that, When the processor executes the computer program on the computer-readable storage medium, it implements the steps of the fall detection method based on video data according to any one of claims 1-8.

Images