Dynamic tracking detection method and device for disabled people

Abstract

Embodiments of the present disclosure disclose a dynamic tracking detection method and device for the disabled. A specific embodiment of the method comprises: acquiring basic information, a set of training guide information and a sequence of work scene images for a target disabled person; performing data standardization on the basic information and the set of training guide information to obtain standardized basic information and a set of standardized training guide information; constructing an evaluation index matrix; in response to determining that the disability type meets a first preset type condition, performing target dynamic behavior detection on each work scene image in the sequence of work scene images by using a preset dynamic behavior detection model to generate a sequence of target behavior detection results; and performing ability assessment on the target disabled person according to the evaluation index matrix and the sequence of target behavior detection results to generate disability ability assessment information. The embodiment can improve the accuracy of the detection results.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the fields of disability assessment technology, target dynamic tracking and detection technology, and computer technology, specifically to dynamic tracking and detection methods and apparatus for people with disabilities. Background Technology

[0002] Disability assessment, through a combination of scientific and systematic training and multi-dimensional ability evaluation, comprehensively and accurately identifies the potential abilities and strengths of individuals with disabilities. Its core objective is to break down information barriers and cognitive biases in the employment process for people with disabilities, enabling each individual to find a job that matches their abilities, fully realize their value, and promote the protection of their employment rights and the improvement of social inclusion. Currently, existing software systems often provide a one-time assessment of an individual's abilities.

[0003] However, the above methods lack a dynamic tracking mechanism (i.e., a long-term, continuous tracking mechanism), making it difficult to detect the target's disability ability and environmental adaptation in a timely manner, resulting in delayed assessment results. Summary of the Invention

[0004] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0005] Some embodiments of this disclosure propose a dynamic tracking and detection method and apparatus for people with disabilities to address the technical problems mentioned in the background section above.

[0006] In a first aspect, some embodiments of this disclosure provide a dynamic tracking and detection method for persons with disabilities. The method includes: acquiring basic information, a training guidance information set, and a sequence of work scene images for the target person with a disability; wherein the basic information includes: disability type, duration of disability, disability level, target educational level, and disability location identifier; standardizing the basic information and the training guidance information set to obtain standardized basic information and a standardized training guidance information set; constructing an evaluation index matrix based on the standardized basic information and the standardized training guidance information set; and, in response to determining that the disability type meets a first preset type condition, performing target dynamic behavior detection on each work scene image in the work scene image sequence using a preset dynamic behavior detection model. The system generates a target behavior detection result sequence, wherein the first preset type condition is used to determine whether the disability type represents sensory impairment; based on the evaluation index matrix and the target behavior detection result sequence, the system performs an ability assessment on the target person with disabilities to generate disability ability assessment information; in response to determining that the disability type meets the second preset type condition, the system performs target posture detection on each work scene image in the work scene image sequence using a preset target posture detection model to generate a target posture detection result sequence, wherein the second preset type condition is used to determine whether the disability type represents activity ability impairment; based on the evaluation index matrix and the target posture detection result sequence, the system performs an ability assessment on the target person with disabilities to generate disability ability assessment information.

[0007] Secondly, some embodiments of this disclosure provide a dynamic tracking and detection device for persons with disabilities. The device includes: an acquisition unit configured to acquire basic information, a training guidance information set, and a sequence of work scene images for the target person with disabilities, wherein the basic information includes: disability type, duration of disability, disability level, target education level, and disability location identifier; a data standardization unit configured to standardize the basic information and the training guidance information set to obtain standardized basic information and a standardized training guidance information set; a construction unit configured to construct an evaluation index matrix based on the standardized basic information and the standardized training guidance information set; and a target dynamic behavior detection unit configured to, in response to determining that the disability type meets a first preset type condition, perform target dynamic behavior detection on each work scene image in the work scene image sequence using a preset dynamic behavior detection model. The system includes a target behavior detection unit to generate a target behavior detection result sequence, wherein the first preset type condition is used to determine whether the disability type represents sensory impairment; a first evaluation unit is configured to perform an ability assessment on the target disabled person based on the evaluation index matrix and the target behavior detection result sequence to generate disability ability assessment information; a target posture detection unit is configured to, in response to determining that the disability type meets the second preset type condition, perform target posture detection on each work scene image in the work scene image sequence using a preset target posture detection model to generate a target posture detection result sequence, wherein the second preset type condition is used to determine whether the disability type represents activity ability impairment; and a second evaluation unit is configured to perform an ability assessment on the target disabled person based on the evaluation index matrix and the target posture detection result sequence to generate disability ability assessment information.

[0008] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0009] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0010] The various embodiments of this disclosure have the following beneficial effects: the dynamic tracking and detection method for persons with disabilities according to some embodiments of this disclosure can improve the accuracy of disability ability detection. Specifically, the dynamic tracking and detection method for persons with disabilities according to some embodiments of this disclosure first acquires basic information, training guidance information set, and work scene image sequence for the target person with disabilities. The basic information includes: disability type, duration of disability, disability level, target education level, and disability location identifier. Here, by introducing the work scene image sequence, the actual working situation of the person with disabilities can be determined in real time. Therefore, the ability of the person with disabilities can be dynamically detected and updated in a timely manner. Then, the basic information and training guidance information set are standardized to obtain standardized basic information and standardized training guidance information set. Afterwards, an evaluation index matrix is ​​constructed based on the standardized basic information and standardized training guidance information set. Here, by constructing the evaluation index matrix, it can be used to conduct a more comprehensive assessment of the ability of persons with disabilities. Next, in response to determining that the aforementioned disability type meets the first preset type condition, a preset dynamic behavior detection model is used to perform target dynamic behavior detection on each work scene image in the aforementioned work scene image sequence to generate a target behavior detection result sequence. The first preset type condition is used to determine whether the disability type represents sensory impairment. Here, by introducing a dynamic behavior detection model, it can be used to detect the dynamic behavior of disabled persons in the work scene, thereby assisting in the assessment of the disabled person's abilities. Then, based on the aforementioned evaluation index matrix and the aforementioned target behavior detection result sequence, the abilities of the aforementioned target disabled persons are assessed to generate disability ability assessment information. Thus, fine-grained ability assessment can be performed on disabled persons under the first preset type condition, realizing dynamic tracking and detection of work abilities. Furthermore, in response to determining that the aforementioned disability type meets the second preset type condition, a preset target posture detection model is used to perform target posture detection on each work scene image in the aforementioned work scene image sequence to generate a target posture detection result sequence. The second preset type condition is used to determine whether the disability type represents a motor ability impairment. Here, considering that different types of disabilities have varying degrees of impact on work and require different work scenarios, a first and second preset type condition are introduced to differentiate between them and avoid feature conflicts caused by simultaneous detection. This avoids feature conflicts and improves detection accuracy. Finally, based on the aforementioned evaluation index matrix and the target posture detection result sequence, the abilities of the target disabled individuals are assessed to generate disability ability assessment information. This avoids the lag in assessment results, allowing for timely detection of the target's disability ability and environmental adaptability, thus improving the accuracy of the detection results. Attached Figure Description

[0011] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0012] Figure 1 This is a flowchart of some embodiments of the dynamic tracking and detection method for people with disabilities according to this disclosure;

[0013] Figure 2 This is a schematic diagram of the overall framework of the human pose recognition model;

[0014] Figure 3 This is a schematic diagram of the structure of some embodiments of the dynamic tracking and detection device for people with disabilities according to the present disclosure;

[0015] Figure 4 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0016] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0017] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0018] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0019] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0020] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0021] Before performing any of the operations involving the collection, storage, and use of user personal information (such as basic information of the target disabled persons, training guidance information sets, and work scene image sequences) in this disclosure, the relevant organizations or individuals shall fulfill their obligations, including conducting personal information security impact assessments, informing the personal information subjects, and obtaining prior authorization and consent from the personal information subjects.

[0022] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] Figure 1 A flow 100 of some embodiments of a dynamic tracking and detection method for persons with disabilities according to the present disclosure is shown. The dynamic tracking and detection method for persons with disabilities includes the following steps:

[0024] Step 101: Obtain basic information, training guidance information set, and work scene image sequence for the target disabled persons.

[0025] In some embodiments, the entity executing the dynamic tracking and detection method for persons with disabilities (e.g., a computing device) can acquire basic information, training guidance information sets, and work scene image sequences for the target person with disabilities via wired or wireless means. The basic information may include: disability type, duration of disability, disability level, target educational level, and disability location identifier. Disability types include: visual impairment, hearing impairment, speech impairment, intellectual disability, physical disability, etc. Disability levels can be pre-defined based on the degree of disability. Here, disability location identifiers can be unique markers pre-made for various parts of the body. The training guidance information set can be obtained after providing job training to the (target) person with disabilities. Each piece of training guidance information can correspond to a training result. For example, training guidance information might correspond to a training course completion rate of 80%. Another example is a training test score. The work scene image sequence can be a series of images taken during the target person's on-the-job period.

[0026] It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra wideband) connections, and other currently known or future wireless connection methods.

[0027] It should be noted that the aforementioned computing devices can be either hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device. When the computing device is software, it can be installed on the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here.

[0028] Optionally, the above training guidance information set is generated through the following steps:

[0029] Step S1: Synchronously acquire the first training image sequence and the second training image sequence. Each first training image corresponds one-to-one with each second training image. The first training image is an image of the target person with disabilities, and the second training image is an image of the instructor.

[0030] Step S2 involves extracting key points from each first training image in the first training image sequence to generate a first key point sequence, and then using these first key point sequences to generate a first topological structure sequence. The BlazePose (human pose tracking) algorithm can be used to extract key points from each first training image in the first training image sequence to generate the first key point sequence. For example, 33 first key points can be extracted. The first key point sequence can represent the body parts of the target person with disabilities. For example, body part key points can include, but are not limited to: eyes, nose, ears, shoulders, elbows, wrists, knees, and ankles. Furthermore, using the body part key point "nose" as the root node, the first topological structure sequence can be generated by sequentially connecting body part key points belonging to the head, body part key points belonging to the torso, and body part key points belonging to the limbs.

[0031] Step S3 involves extracting key points from each second training image in the second training image sequence to generate a second key point sequence, and then using these second key point sequences to generate a second topology sequence. The BlazePose (human pose tracking) algorithm can be used to extract key points from each second training image in the second training image sequence to generate the second key point sequence. Furthermore, the steps for generating the second topology sequence can refer to the implementation method of the first topology sequence described above, and will not be elaborated further.

[0032] Step S4: Perform genetic matching on the first topology sequence and the second topology sequence to obtain the optimized pose parameter set sequence. The optimized pose parameter set sequence can be obtained by using a genetic algorithm to perform genetic matching on the first topology sequence and the second topology sequence.

[0033] Step S5: Using the optimized pose parameter set sequence described above, a training guidance information set is generated. This set can also include a training guidance map to guide individuals with disabilities in continuing training, allowing for the capture of more training images to generate further training guidance information.

[0034] Optionally, a preset human pose recognition model can be used to identify the first training image sequence and the second training image sequence to obtain the aforementioned first topological structure sequence and the aforementioned second topological structure sequence. Then, genetic matching is performed on the aforementioned first topological structure sequence and the aforementioned second topological structure sequence to obtain the optimized pose parameter set sequence.

[0035] As an example, see Figure 2 The diagram shows the overall framework of the human pose recognition model. The aforementioned human pose recognition model may include a pose extraction module 203, a genetic matching module 204, and a visualization feedback module 205. The pose extraction module 203 may include: a human topology module, a pose tracker network, a body pose detector, and a real-time human tracking inference channel. Here, the human topology module can store pre-constructed human topologies of trainers and instructors. It can also utilize the identified first and second topology sequences to input the human topology into the pose tracker network as structural constraints, generating a temporally smooth 3D keypoint sequence.

[0036] Specifically, the pose tracker network can include a single-frame detection submodule, a temporal correlation submodule, and a smoothing optimization submodule. Here, the input to the single-frame detection submodule can be a single first training image (e.g., an RGB image with a size of 256×256) from the first training image sequence 201. Then, keypoint extraction is performed on the first training image through a backbone network (lightweight convolutional layers, depthwise separable convolutions of MobileNetV2) to obtain the coordinates of multiple 3D keypoints corresponding to the human pose in a single frame and their corresponding confidence scores. Next, the temporal correlation submodule matches keypoints in the current frame with those in the previous frame using a Hungarian algorithm and predicts the possible positions of keypoints in the next frame using a Kalman filter, reducing detection latency. When keypoints are occluded, interpolation is performed based on the historical trajectory of the keypoints and human kinematic constraints (such as joint angle ranges). The smoothing optimization submodule can perform a weighted average of the keypoint coordinates across multiple consecutive frames (e.g., 5 frames). Simultaneously, erroneous detection results are filtered based on confidence level and motion speed threshold (e.g., joint movement speed does not exceed 0.5m / s). Finally, unreasonable key point positions are corrected using the human skeleton connection relationship (e.g., the angle range of the shoulder, elbow, and wrist), to obtain the coordinates of multiple human pose 3D key points and their corresponding confidence levels for a single frame image.

[0037] To address the issue of poor matching between the second and first training videos due to differences in size and shooting position, this application applies normalization and centering processing to the extracted source and target pose 3D key point coordinates, converting the key point coordinates into coordinates relative to the centroid, thus eliminating the effects of translation and scaling.

[0038] Secondly, keypoint detection algorithms can be used as body pose detectors. For example, keypoint detection algorithms can include HRNet (High-Resolution Net) or BlazePose. The input to the body pose detector is a single second training image in the second training image sequence 202, generating multiple 3D keypoint coordinates and corresponding confidence scores for each frame of the second training image. The real-time human tracking inference channel can include a dual-path feature processing layer, a 3D tracking layer, and a pose difference calculation layer. Here, the 33 3D keypoints (99-dimensional) output by the dual-path feature processing layer, the pose tracker network, and the body pose detector are directly compressed into 64-dimensional features through a "1-layer fully connected + ReLU linear rectified function". Thus, the same feature extraction function is used (without parameter sharing / freezing operations, making training and inference simpler). Trunk center calculation: The average 3D coordinates of the original four points "chest, waist, and pelvis (left and right)" are retained as tracking anchor points (ensuring that the anti-occlusion ability remains unchanged). Two 64-dimensional features and two 3D centers are output. Then, two 64-dimensional features and two 3D centers are input into the 3D Kalman filter tracker of the 3D tracking layer to obtain a 6D state vector for each keypoint coordinate. The 6D state vector can include position and velocity components on the horizontal, vertical, and axial axes. The attitude difference calculation layer can generate two types of difference features, including attitude offset features and attitude association features. Attitude offset features: 64-dimensional features of the disabled person - 64-dimensional features of the teacher (used to reflect the "absolute offset" of the two roles' attitudes, such as the feature difference that the disabled person's left shoulder is higher than the teacher's left shoulder); Attitude association features: 64-dimensional features of the disabled person × 64-dimensional features of the teacher (used to reflect the "relative association" of the two roles' attitudes, such as the feature synergy when the two roles raise their hands simultaneously). Finally, the "64-dimensional features of the disabled person + 64-dimensional features of the teacher + attitude offset features (64-dimensional) + attitude association features (64-dimensional)" are concatenated into a 256-dimensional joint feature, input into a 1-layer fully connected fusion network, and output as a 128-dimensional two-role association feature after passing through a normalization layer and a linear rectified function.

[0039] The genetic matching module 204 can take as input the dual-role association features output from the real-time human tracking inference channel mentioned above, and output pose parameters between the pose of the person with disabilities in the first training image and the pose of the instructor in the second training image. These parameters describe the rotation and translation transformation relationship of the person with disabilities' pose relative to the standard pose, i.e., the relative pose matrix. Finally, the relative pose matrix and the corresponding pose description information, training progress, and other data can be used as training guidance information.

[0040] Furthermore, the visualization feedback module 205 can overlay the aligned pose (matrix) with the instructor's standard pose on the same screen based on the relative pose, allowing for a direct comparison of movement differences. It renders the 3D pose using a "skeleton lines + key points" format, with key points marked by colored dots (e.g., red for the head, blue for the torso, and green for the limbs). Skeleton lines are connected by line segments (e.g., blue for the trainee's skeleton and red for the instructor's), resulting in a training guidance diagram that avoids visual clutter. This allows for visual presentation and provides feedback guidance to the trainee.

[0041] Step 102: Standardize the basic information and training guidance information sets to obtain standardized basic information and standardized training guidance information sets.

[0042] In some embodiments, the aforementioned implementing entity may standardize the aforementioned basic information and training guidance information set in various ways to obtain standardized basic information and standardized training guidance information set.

[0043] In some optional implementations of certain embodiments, the executing entity performs data standardization on the aforementioned basic information and the aforementioned training guidance information set to obtain standardized basic information and standardized training guidance information set, including:

[0044] Step S1: Perform numerical mapping on the above basic information to obtain a mapped data set. Specifically, the numerical values ​​in the basic information (e.g., years of disability, target age, etc.) can be normalized to map the values ​​to the [0, 1] interval, eliminating dimensional differences and obtaining the mapped data set.

[0045] Step S2 involves encoding the aforementioned basic information to obtain an information encoding group. Specifically, the disability type, target educational level, and disability location identifier within the basic information can be encoded into numerical codes according to a preset attribute order to obtain the information encoding group.

[0046] Step S3: Determine the above-mentioned mapping data group and the above-mentioned information encoding group as standardized basic information.

[0047] Step S4 involves outlier handling of the training guidance information set to generate a standardized training guidance information set. This can be achieved by using box plots to identify outlier data (such as test scores significantly deviating from the normal range), verifying and correcting outliers, and removing those that cannot be corrected. Secondly, for missing values, mean imputation (suitable for quantitative data, such as training duration) or mode imputation (suitable for qualitative data, such as disability type) can be used to obtain a standardized training guidance information set, ensuring data integrity.

[0048] Step 103: Construct an evaluation index matrix based on standardized basic information and standardized training guidance information set.

[0049] In some embodiments, the implementing entity may construct an evaluation index matrix based on the aforementioned standardized basic information and the aforementioned standardized training guidance information set.

[0050] In some optional implementations of certain embodiments, the executing entity constructs an evaluation index matrix based on the aforementioned standardized basic information and the aforementioned standardized training guidance information set, including:

[0051] Step S1: Determine the indicator attributes corresponding to the aforementioned standardized basic information and standardized training guidance information set to obtain an indicator attribute sequence. Specifically, attributes from the standardized basic information and standardized training guidance information can be extracted as indicator attributes using an index to obtain the indicator attribute sequence.

[0052] Step S2: According to the preset index arrangement order, construct an evaluation index matrix by combining the attribute values ​​corresponding to each index attribute in the above index attribute sequence. This can be achieved by combining the various index attributes into an attribute matrix according to the preset arrangement order. Then, fill the corresponding attribute values ​​of the index attributes into the corresponding positions in the attribute matrix to obtain the evaluation index matrix.

[0053] As an example, the indicator attribute sequence includes 16 indicator attributes, which can be used to construct a 4×4 evaluation indicator matrix.

[0054] Step 104: In response to determining that the disability type meets the first preset type condition, target dynamic behavior detection is performed on each work scene image in the work scene image sequence through a preset dynamic behavior detection model to generate a target behavior detection result sequence.

[0055] In some embodiments, the execution entity may, in response to determining that the disability type meets a first preset type condition, perform target dynamic behavior detection on each work scene image in the work scene image sequence using a preset dynamic behavior detection model to generate a target behavior detection result sequence. The first preset type condition is used to determine whether the disability type represents sensory impairment.

[0056] In practice, considering the varying impacts of different types of disability on abilities, such as the hearing and language communication impairments faced by deaf-mute individuals and the visual perception deficits of blind individuals, while motor disability primarily affects an individual's physical motor function—for example, individuals with limb amputations may have limitations in the integrity and flexibility of their limb movements, and patients with cerebral palsy may experience problems with motor coordination and control. Because these two types of disabilities result in completely different functional impairments, they need to be assessed separately to accurately grasp the core issues faced by individuals with different disabilities. Furthermore, individuals with sensory impairments often exhibit dependence on or lack of specific sensory information; their abilities are more reflected in the level of compensation through other senses and their ability to utilize special communication and assistive technologies. For example, deaf-mute individuals may excel at communicating and completing tasks through sign language and visual observation, while blind individuals may have high abilities in auditory perception and tactile discrimination. The behavioral manifestations of individuals with motor disability mainly focus on the completion of bodily movements, the stability and flexibility of movement. For example, the performance of individuals with limb disabilities in operating tools, moving their bodies, and their range of motion and muscle strength. Therefore, different assessment methods need to be designed for different behavioral manifestations and ability demonstrations. Thus, this application classifies disability types into sensory impairment (i.e., disability types that meet the first preset type conditions) and motor impairment (i.e., disability types that meet the second preset type conditions). This classification can be used to differentiate between sensory impairment and motor impairment for a comprehensive and accurate assessment of the abilities of persons with disabilities. This allows for targeted dynamic tracking and improves the accuracy of disability ability detection.

[0057] In some optional implementations of certain embodiments, the dynamic behavior detection model includes: an input layer, a spatiotemporal feature extraction network, a behavior feature optimization network, and a behavior classification module. The execution entity uses the preset dynamic behavior detection model to perform target dynamic behavior detection on each work scene image in the above-mentioned work scene image sequence to generate a target behavior detection result sequence, including:

[0058] Step S1: Through the input layer described above, image processing is performed on each work scene image in the work scene image sequence to obtain a first image sequence. Specifically, the input layer can sequentially extract a target number of work scene images (e.g., 100 frames) from the work scene image sequence as the first image sequence. Here, the first image can be an RGB three-channel image with a resolution of 1920×1080. Additionally, the input layer can also employ bilateral filtering algorithms to denoise the work scene images and perform pixel normalization (linearly transforming pixel values ​​from [0, 255] to the [0, 1] interval), among other operations.

[0059] Step S2: Using the aforementioned spatiotemporal feature extraction network, spatiotemporal features are extracted from the first image sequence to obtain a temporal attention feature vector. The spatiotemporal feature extraction network can be composed of a 3D convolutional neural network and a temporal attention mechanism.

[0060] As an example, a 3D convolutional neural network can include four convolutional blocks, each consisting of a 3D convolutional layer, a BatchNorm layer, a LeakyReLU activation function, and a 3D max-pooling layer. The temporal attention mechanism then converts the feature map (1×10×10×512) output from the 3D convolutional layer into a temporal feature vector (1×10×51200). Attention weights are calculated at each time step using a fully connected layer, and after Softmax normalization, a weighted sum of these weights yields a 1×51200 temporal attention feature vector.

[0061] Step S3: The temporal attention feature vector is optimized using the aforementioned behavior feature optimization network to obtain an optimized feature vector. The behavior feature optimization network can include a feature optimization module and a behavior classification module. Specifically, the feature optimization module can concatenate the temporal attention feature vector (1×51200) with the disability type information encoding (1×3) and the standardized value of the basic skill level (1×1), adding them to form a fused feature vector of 1×51204. Next, after dimensionality reduction using two layers of MLP (Multilayer Perceptron) (input 51204, output 1024, activation function ReLU for both), a 1×1024 optimized feature vector is obtained.

[0062] Step S4: The optimized feature vector is classified into behaviors using the behavior classification module to obtain the target behavior detection result sequence. The behavior classification module uses a two-layer fully connected network (first layer input 1024 → output 512; second layer input 512 → output K, where K is the number of behavior categories, e.g., K = 5). Then, a Softmax classifier outputs the probabilities of each behavior category to obtain the target behavior detection result sequence.

[0063] Step 105: Based on the evaluation index matrix and the target behavior detection result sequence, conduct an ability assessment on the target person with disabilities to generate disability ability assessment information.

[0064] In some embodiments, the aforementioned implementing entity may conduct a capability assessment of the aforementioned target person with disabilities based on the aforementioned evaluation index matrix and the aforementioned target behavior detection result sequence, so as to generate disability capability assessment information.

[0065] In some optional implementations of certain embodiments, the executing entity performs a capability assessment on the target person with disabilities based on the evaluation index matrix and the target behavior detection result sequence, to generate disability capability assessment information, including:

[0066] Step S1 involves performing feature analysis on each target behavior detection result in the aforementioned target behavior detection result sequence to generate a feature analysis dataset. Feature analysis may involve determining features such as duration and behavior confidence for each behavior category in each target behavior detection result.

[0067] As an example, the feature analysis data includes: behavior category (e.g., "complete typing task"), behavior confidence (e.g., 0.92), and behavior duration (30 minutes).

[0068] Step S2: Using the aforementioned feature analysis dataset, the evaluation index matrix is ​​modified to obtain the target index matrix. Here, the evaluation index values ​​corresponding to the feature analysis data in the evaluation index matrix can be adjusted sequentially.

[0069] For example, if a person with a disability has a "test score improvement rate" of 30%, and the mean of all samples for this indicator is 20% and the standard deviation is 10%, then the standardized value is 1.

[0070] Step S3 involves constructing a judgment matrix and performing a consistency check on it. First, the importance of each indicator can be compared pairwise based on expert experience, and the corresponding importance scores can be used as matrix data to generate the judgment matrix. Second, a consistency check algorithm can be used to calculate the consistency ratio of the matrix. If the consistency ratio is less than 0.1, the judgment matrix has passed the consistency check. Otherwise, the judgment matrix is ​​readjusted.

[0071] Step S4: In response to the successful consistency check, determine the eigenvalue matrix corresponding to the largest eigenvalue of the judgment matrix, and use it as the weight matrix. This weight matrix can be obtained by using the eigenvalue method to identify the eigenvalue matrix corresponding to the largest eigenvalue in the judgment matrix.

[0072] Step S5: Based on the aforementioned weight matrix and target indicator matrix, generate basic competency indicator scores, training indicator scores, and work indicator scores. Specifically, the basic competency indicator scores are obtained by weighting the evaluation indicator values ​​belonging to basic competencies in the target indicator matrix with their corresponding weights in the weight matrix. The training indicator scores are obtained by weighting the evaluation indicator values ​​belonging to training indicators with their corresponding weights in the weight matrix. The work indicator scores are obtained by weighting the evaluation indicator values ​​belonging to work indicators with their corresponding weights in the weight matrix.

[0073] Step S6: Weight the scores of the above-mentioned basic ability indicators, training indicators, and work indicators to generate a disability ability score, which serves as disability ability assessment information.

[0074] As an example, the scores of basic competency indicators, training effectiveness indicators, and job fit indicators can be weighted and summed according to a weight of 3:3:4 to obtain a disability competency score, which can be used as disability competency assessment information.

[0075] Step 106: In response to determining that the disability type meets the second preset type condition, target pose detection is performed on each work scene image in the work scene image sequence using a preset target pose detection model to generate a target pose detection result sequence.

[0076] In some embodiments, the execution entity may, in response to determining that the disability type meets a second preset type condition, perform target pose detection on each work scene image in the work scene image sequence using a preset target pose detection model to generate a target pose detection result sequence. The second preset type condition is used to determine whether the disability type represents a disability in terms of activity ability.

[0077] In some optional implementations of certain embodiments, the target pose detection model includes: an input layer, a pose feature extraction network, a node parameter calculation network, and a pose evaluation layer. The execution entity uses the preset target pose detection model to perform target pose detection on each work scene image in the above-mentioned work scene image sequence to generate a target pose detection result sequence, including:

[0078] Step S1: Through the aforementioned input layer, image processing is performed on each work scene image in the aforementioned work scene image sequence to obtain a second image sequence. The implementation method of the aforementioned input layer can refer to the aforementioned dynamic behavior detection model including an input layer, and will not be described in detail here.

[0079] Step S2: Using the aforementioned pose feature extraction network, pose features are extracted from the second image sequence to obtain a node feature heatmap sequence. The pose feature extraction network can include a backbone network, a feature pyramid module, a prediction branch, and a joint feature enhancement module. Specifically, the backbone network can be a MobileNetV3-Large model, outputting feature maps at five scales. The feature pyramid module can fuse the five scale feature maps to generate three fused feature maps. The prediction branch can be used to predict joint heatmaps and correlation vector maps. Three-scale joint heatmaps (640×640×33) are output and weighted fusion is performed to obtain the node feature heatmap. Three-scale correlation vector maps (640×640×64) are output. 33 keypoints correspond to 32 pairs of adjacent joints (e.g., top of head-brow arch, brow arch-nose tip, nose tip-chin, shoulder joint-elbow joint, etc.). Each pair of joints requires two channels to describe the x-axis and y-axis relationship, 32×2=64, which are then weighted and fused to obtain the correlation vector map. Here, generating an association vector graph can be used to ensure that the coordinates of the same joint are traceable in consecutive frames. At the same time, if the joint coordinates shift significantly in a frame, the shifted coordinates can be corrected by maintaining the consistency of the direction of the association vectors in adjacent frames (e.g., when the knee joint coordinates suddenly shift, the direction of the association vector from the hip joint to the knee joint can be combined to adjust it to a reasonable position), thus avoiding the flexibility calculation error caused by inter-frame jitter.

[0080] In practice, the calculation of joint flexibility needs to be based on the "motion trajectory of the same joint in consecutive frames", and the posture stability needs to be based on the "coordinate standard deviation of normal joints". If the associated vector map is not generated, joint mismatch will occur, resulting in "cross-joint calculation trajectory" (such as splicing the coordinates of the left knee joint in frame 1 with the coordinates of the right knee joint in frame 2), making the flexibility and stability values ​​completely lose their physical meaning.

[0081] Joint Feature Enhancement Module: The input is a one-hot encoded vector of the disability type (1×3, such as left upper limb amputation = 0, right lower limb amputation = 1, both lower limbs amputation = 2), which is multiplied element by element with the joint heatmap features. The features of missing joints (such as the left elbow joint and left wrist joint of a person with a missing left upper limb) are set to 0, while the features of normal joints are retained.

[0082] Step S3: Using the node parameter calculation network described above, the node parameters of each node feature heatmap in the node feature heatmap sequence are determined, resulting in a node parameter sequence. The node parameter calculation network can be used to locate joint coordinates, calculate joint pressure, and flexibility. Specifically, the maximum coordinate (x, y) of each joint heatmap is taken as the position of that joint in the image; if a joint is missing (the maximum heatmap value is 0), it is marked as "N". Next, taking the joint coordinates (x, y) as the center, an 11×11 pixel local region is defined, and the Sobel operator is used to calculate the gray-level gradients in the x and y directions of this region, obtaining the gradient magnitude matrix. Then, the mean gradient magnitude G is calculated and standardized according to the formula "Joint pressure value = G / Gmax" (Gmax is the maximum mean gradient of the same joint in a normal population, obtained through statistical experiments in normal population work scenarios), with the result ∈ [0, 1]. The pressure value of a missing joint is recorded as 0.

[0083] The joint flexibility of each joint can be calculated using the following steps: First, extract the coordinates of the same joint from 10 consecutive frames in the image sequence and calculate the coordinate distance between adjacent frames. Then, calculate the average movement distance based on the coordinate distance between adjacent frames. Next, take the joint coordinates of 3 consecutive frames and calculate the joint angle using the vector dot product formula. Finally, calculate the joint flexibility using the following formula: Joint flexibility = 0.6 × (average movement distance / maximum average joint movement distance) + 0.4 × (joint angle / maximum joint range of motion). Therefore, the joint pressure value and joint flexibility of each joint can be used as the corresponding node parameters.

[0084] As an example, node parameters may include: shoulder joint coordinates (x = 120, y = 180), knee joint pressure value: 35 N, or elbow joint flexibility: 0.75.

[0085] Step S4: Using the aforementioned attitude evaluation layer and the aforementioned node parameter sequence, attitude evaluation is performed on the feature heatmaps of each node in the aforementioned node feature heatmap sequence to generate target attitude detection results, resulting in a target attitude detection result sequence. The attitude evaluation may involve assessing the joint stress risk of each joint.

[0086] As an example, joint stress risk assessment could involve counting the number of joints in each frame whose stress values ​​exceed a preset threshold. If the number exceeds the preset threshold, a stress assessment result representing the stress risk of the joint is generated.

[0087] In practice, individuals with mobility impairments often experience excessive stress on their joints due to incomplete limb structures (e.g., those with upper limb amputation rely on unilateral limb exertion, while those with lower limb amputation rely on prostheses). This stress can accumulate over time and lead to irreversible damage. Therefore, joint stress assessment can further determine the stress level of each joint, enabling timely risk warnings and preventing irreversible joint damage.

[0088] Step 107: Based on the evaluation index matrix and the target posture detection result sequence, conduct a capability assessment of the target person with disabilities to generate disability capability assessment information.

[0089] In some embodiments, the executing entity may assess the capabilities of the target person with disabilities based on the evaluation index matrix and the target posture detection result sequence to generate disability assessment information. The implementation method and corresponding technical effects of this step can be referenced from the implementation of step 105, and will not be described in detail hereafter.

[0090] Optionally, the aforementioned implementing entity may also perform the following steps:

[0091] Step S1: In response to determining that the disability type satisfies the first preset type condition and the second preset type condition, the dynamic behavior detection model and the target posture detection model are used to track and detect each work scene image in the work scene image sequence to generate a target tracking and detection result sequence. The disability type satisfying the first and second preset type conditions indicates that the person with the disability has both sensory impairment and limited mobility. Therefore, the dynamic behavior detection model and the target posture detection model can be used simultaneously for tracking and detection. Here, the detection results of the dynamic behavior detection model and the target posture detection model can be weighted and summed to obtain the target tracking and detection result sequence.

[0092] Step S2 involves assessing the abilities of the target disabled person based on the aforementioned evaluation index matrix and target tracking detection result sequence, thereby generating disability assessment information. The implementation method and corresponding technical effects of this step can be referenced from the implementation method of step 105 above, and will not be elaborated further.

[0093] Step S3: Based on the disability assessment information, generate employment guidance information and send it to the target terminal. If the disability assessment information includes information characterizing work risk (e.g., job adaptability below an adaptability threshold, node stress value above a preset stress threshold), then the disability assessment information is sent to a training guidance terminal to generate employment guidance information. Here, the training guidance terminal can be a user terminal for training and guiding people with disabilities. The employment guidance information can be information for adjusting the job position or training direction of people with disabilities. The target terminal can include the smart terminal of the person with disabilities and the aforementioned training guidance terminal.

[0094] The various embodiments of this disclosure have the following beneficial effects: the dynamic tracking and detection method for persons with disabilities according to some embodiments of this disclosure can improve the accuracy of disability ability detection. Specifically, the dynamic tracking and detection method for persons with disabilities according to some embodiments of this disclosure first acquires basic information, training guidance information set, and work scene image sequence for the target person with disabilities. The basic information includes: disability type, duration of disability, disability level, target education level, and disability location identifier. Here, by introducing the work scene image sequence, the actual working situation of the person with disabilities can be determined in real time. Therefore, the ability of the person with disabilities can be dynamically detected and updated in a timely manner. Then, the basic information and training guidance information set are standardized to obtain standardized basic information and standardized training guidance information set. Afterwards, an evaluation index matrix is ​​constructed based on the standardized basic information and standardized training guidance information set. Here, by constructing the evaluation index matrix, it can be used to conduct a more comprehensive assessment of the ability of persons with disabilities. Next, in response to determining that the aforementioned disability type meets the first preset type condition, a preset dynamic behavior detection model is used to perform target dynamic behavior detection on each work scene image in the aforementioned work scene image sequence to generate a target behavior detection result sequence. The first preset type condition is used to determine whether the disability type represents sensory impairment. Here, by introducing a dynamic behavior detection model, it can be used to detect the dynamic behavior of disabled persons in the work scene, thereby assisting in the assessment of the disabled person's abilities. Then, based on the aforementioned evaluation index matrix and the aforementioned target behavior detection result sequence, the abilities of the aforementioned target disabled persons are assessed to generate disability ability assessment information. Thus, fine-grained ability assessment can be performed on disabled persons under the first preset type condition, realizing dynamic tracking and detection of work abilities. Furthermore, in response to determining that the aforementioned disability type meets the second preset type condition, a preset target posture detection model is used to perform target posture detection on each work scene image in the aforementioned work scene image sequence to generate a target posture detection result sequence. The second preset type condition is used to determine whether the disability type represents a motor ability impairment. Here, considering that different types of disabilities have varying degrees of impact on work and require different work scenarios, a first and second preset type condition are introduced to differentiate between them and avoid feature conflicts caused by simultaneous detection. This avoids feature conflicts and improves detection accuracy. Finally, based on the aforementioned evaluation index matrix and the target posture detection result sequence, the abilities of the target disabled individuals are assessed to generate disability ability assessment information. This avoids the lag in assessment results, allowing for timely detection of the target's disability ability and environmental adaptability, thus improving the accuracy of the detection results.

[0095] Further reference Figure 3As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a dynamic tracking and detection device for people with disabilities. These device embodiments are similar to... Figure 3 Corresponding to the method embodiments shown, this dynamic tracking and detection device for people with disabilities can be specifically applied to various electronic devices.

[0096] like Figure 3 As shown, a dynamic tracking and detection device 300 for persons with disabilities in some embodiments includes: an acquisition unit 301, a data standardization unit 302, a construction unit 303, a target dynamic behavior detection unit 304, a first evaluation unit 305, a target posture detection unit 306, and a second evaluation unit 307. The acquisition unit 301 is configured to acquire basic information, a training guidance information set, and a sequence of work scene images for the target person with disabilities. The basic information includes: disability type, duration of disability, disability level, target education level, and disability location identifier. The data standardization unit 302 is configured to standardize the basic information and the training guidance information set to obtain standardized basic information and a standardized training guidance information set. The construction unit 303 is configured to construct an evaluation index matrix based on the standardized basic information and the standardized training guidance information set. The target dynamic behavior detection unit 304 is configured to, in response to determining that the disability type meets a first preset type condition, perform target dynamic behavior detection on each work scene image in the work scene image sequence using a preset dynamic behavior detection model to generate a target behavior detection result sequence. The system comprises: a first preset type condition for determining whether a disability type represents a sensory disability; a first assessment unit 305 configured to assess the abilities of the target person with a disability based on the evaluation index matrix and the target behavior detection result sequence, thereby generating disability ability assessment information; a target posture detection unit 306 configured to, in response to determining that the disability type meets the second preset type condition, perform target posture detection on each work scene image in the work scene image sequence using a preset target posture detection model, thereby generating a target posture detection result sequence, wherein the second preset type condition is used to determine whether a disability type represents a motor ability disability; and a second assessment unit 307 configured to assess the abilities of the target person with a disability based on the evaluation index matrix and the target posture detection result sequence, thereby generating disability ability assessment information.

[0097] It is understandable that the units described in the dynamic tracking and detection device 300 for people with disabilities are similar to those in the reference device. Figure 1 The steps in the described method correspond accordingly. Therefore, the operations, features, and beneficial effects described above for the method are also applicable to the dynamic tracking and detection device 300 for people with disabilities and the units contained therein, and will not be repeated here.

[0098] The following is for reference. Figure 4 It illustrates a schematic diagram of the structure of an electronic device (such as a computing device) suitable for implementing some embodiments of the present disclosure. Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality or scope of the embodiments of this disclosure. Figure 4 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The memory may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor to perform any of the methods described above. The processor provides computational and control capabilities to support the operation of the entire computer device. The internal memory provides an environment for the execution of the computer program in the non-volatile storage medium; when executed by the processor, the computer program causes the processor to perform any of the methods described above. The network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the computer device to which the present disclosure is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0099] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be 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. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0100] In one embodiment, the processor is configured to run a computer program stored in a memory to perform the following steps: acquiring basic information, a training guidance information set, and a sequence of work scene images for the target person with disabilities, wherein the basic information includes: disability type, duration of disability, disability level, target education level, and disability location identifier; standardizing the basic information and the training guidance information set to obtain standardized basic information and a standardized training guidance information set; constructing an evaluation index matrix based on the standardized basic information and the standardized training guidance information set; and, in response to determining that the disability type meets a first preset type condition, performing target dynamic behavior detection on each work scene image in the sequence of work scene images using a preset dynamic behavior detection model. The system generates a target behavior detection result sequence, wherein the first preset type condition is used to determine whether the disability type represents sensory impairment; based on the evaluation index matrix and the target behavior detection result sequence, the system performs an ability assessment on the target person with disabilities to generate disability ability assessment information; in response to determining that the disability type meets the second preset type condition, the system performs target posture detection on each work scene image in the work scene image sequence using a preset target posture detection model to generate a target posture detection result sequence, wherein the second preset type condition is used to determine whether the disability type represents activity ability impairment; based on the evaluation index matrix and the target posture detection result sequence, the system performs an ability assessment on the target person with disabilities to generate disability ability assessment information.

[0101] This disclosure also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, and the method implemented when the program instructions are executed can be referred to the various embodiments of the methods described above.

[0102] The aforementioned computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. Alternatively, the aforementioned computer-readable storage medium may be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0103] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0104] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A dynamic tracking and detection method for people with disabilities, characterized in that, include: Acquire basic information, training guidance information set and work scene image sequence for the target disabled persons, wherein the basic information includes: disability type, duration of disability, disability level, target education level and disability location identifier; The basic information and the training guidance information set are standardized to obtain standardized basic information and standardized training guidance information set; Based on the standardized basic information and the standardized training guidance information set, an evaluation index matrix is ​​constructed. In response to determining that the disability type meets the first preset type condition, a target dynamic behavior detection is performed on each work scene image in the work scene image sequence using a preset dynamic behavior detection model to generate a target behavior detection result sequence, wherein the first preset type condition is used to determine whether the disability type represents sensory disability. Based on the evaluation index matrix and the target behavior detection result sequence, the ability of the target person with disabilities is assessed to generate disability ability assessment information; In response to determining that the disability type meets the second preset type condition, the target pose detection is performed on each work scene image in the work scene image sequence through a preset target pose detection model to generate a target pose detection result sequence, wherein the second preset type condition is used to determine whether the disability type represents a disability in mobility. Based on the evaluation index matrix and the target posture detection result sequence, the target person with disabilities is assessed to generate disability assessment information.

2. The method according to claim 1, characterized in that, The method further includes: In response to determining that the disability type satisfies the first preset type condition and the second preset type condition, the dynamic behavior detection model and the target pose detection model are used to track and detect each work scene image in the work scene image sequence to generate a target tracking and detection result sequence. Based on the evaluation index matrix and the target tracking and detection result sequence, the ability of the target person with disabilities is assessed to generate disability ability assessment information. Based on the disability assessment information, employment guidance information is generated and sent to the target terminal.

3. The method according to claim 1, characterized in that, The process of standardizing the basic information and the training guidance information set to obtain standardized basic information and standardized training guidance information set includes: The basic information is numerically mapped to obtain a mapped data set; The basic information is encoded to obtain an information encoding group; The mapping data group and the information encoding group are identified as standardized basic information; Outlier handling is performed on the training guidance information set to generate a standardized training guidance information set.

4. The method according to claim 1, characterized in that, The evaluation index matrix is ​​constructed based on the standardized basic information and the standardized training guidance information set, including: Determine the indicator attributes corresponding to the standardized basic information and the standardized training guidance information set to obtain the indicator attribute sequence; According to the preset index arrangement order, the attribute values ​​corresponding to each index attribute in the index attribute sequence are used to construct an evaluation index matrix.

5. The method according to claim 1, characterized in that, The training guidance information set is generated through the following steps: The first training image sequence and the second training image sequence are acquired simultaneously, wherein the first training image is an image of the target person with disabilities, and the second training image is an image of the instructor. Key points are extracted from each first training image in the first training image sequence to generate a first key point sequence, and a first topological structure sequence is generated using the obtained first key point sequences. Keypoints are extracted from each of the second training images in the second training image sequence to generate a second keypoint sequence, and a second topology sequence is generated using the obtained second keypoint sequences. Genetic matching is performed on the first topological structure sequence and the second topological structure sequence to obtain the optimized pose parameter set sequence; The optimized pose parameter set is used to generate a training guidance information set.

6. The method according to claim 1, characterized in that, The step of assessing the abilities of the target person with disabilities based on the evaluation index matrix and the target behavior detection result sequence to generate disability ability assessment information includes: Feature analysis is performed on each target behavior detection result in the target behavior detection result sequence to generate a feature analysis dataset; Using the feature analysis dataset, the evaluation index matrix is ​​corrected to obtain the target index matrix; Construct a judgment matrix and perform a consistency check on the judgment matrix; In response to the successful consistency check, the feature matrix corresponding to the largest eigenvalue of the judgment matrix is ​​determined and used as the weight matrix. Based on the weight matrix and the target indicator matrix, basic capability indicator scores, training indicator scores, and work indicator scores are generated. The scores for the basic competency indicators, the training indicators, and the work indicators are weighted to generate a disability competency score, which serves as disability competency assessment information.

7. The method according to claim 3, characterized in that, The dynamic behavior detection model includes an input layer, a spatiotemporal feature extraction network, a behavior feature optimization network, and a behavior classification module. The step of using the preset dynamic behavior detection model to perform target dynamic behavior detection on each work scene image in the work scene image sequence to generate a target behavior detection result sequence includes: The input layer performs image processing on each work scene image in the work scene image sequence to obtain a first image sequence. The spatiotemporal feature extraction network is used to extract spatiotemporal features from the first image sequence to obtain a temporal attention feature vector. The temporal attention feature vector is optimized by using the behavioral feature optimization network to obtain an optimized feature vector. The behavior classification module performs behavior classification on the optimized feature vector to obtain a sequence of target behavior detection results.

8. A dynamic tracking and detection device for people with disabilities, characterized in that, include: The acquisition unit is configured to acquire basic information, training guidance information set and work scene image sequence for the target person with disabilities, wherein the basic information includes: disability type, duration of disability, disability level, target education level and disability location identifier; The data standardization unit is configured to perform data standardization on the basic information and the training guidance information set to obtain standardized basic information and standardized training guidance information set. The construction unit is configured to construct an evaluation index matrix based on the standardized basic information and the standardized training guidance information set; The target dynamic behavior detection unit is configured to, in response to determining that the disability type meets a first preset type condition, perform target dynamic behavior detection on each work scene image in the work scene image sequence through a preset dynamic behavior detection model to generate a target behavior detection result sequence, wherein the first preset type condition is used to determine whether the disability type represents sensory disability. The first assessment unit is configured to assess the abilities of the target person with disabilities based on the evaluation index matrix and the target behavior detection result sequence, so as to generate disability ability assessment information. The target pose detection unit is configured to, in response to determining that the disability type meets the second preset type condition, perform target pose detection on each work scene image in the work scene image sequence through a preset target pose detection model to generate a target pose detection result sequence, wherein the second preset type condition is used to determine whether the disability type represents a disability in mobility. The second assessment unit is configured to assess the abilities of the target person with disabilities based on the evaluation index matrix and the target posture detection result sequence, so as to generate disability ability assessment information.

9. An electronic device, characterized in that, include: One or more processors; Storage device, on which one or more programs are stored, When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

10. A computer-readable medium, characterized in that, It stores a computer program thereon, wherein the program, when executed by a processor, implements the method as described in any one of claims 1-7.

Images