Image evaluation method and system for daily activity ability of home-based elderly people

Abstract

This invention relates to the field of image analysis technology, specifically to a method and system for assessing the daily activity capabilities of elderly people living at home. The method involves acquiring continuous temporal images at home and constructing a physical boundary model, extracting the three-dimensional coordinates of the human centroid and upper limb extremities. Combining the shortest spatial distance and dynamic cooperative sliding features, the continuous temporal sequence is divided into effective supported frames and unsupported free frames. The supported frame rate and trajectory bias cost are fused to obtain an initial environmental support feature value. If this value exceeds a threshold, the variance of the free frame centroid fluctuation is extracted and mapped to gait stability confidence. Combined with the support time continuity index, a non-monotonic inversion mechanism is triggered to dynamically adjust the fusion weights, thereby generating a corrected environmental support feature value. Finally, this value is compared with a historical baseline to output a decline warning. This method effectively removes pseudo-compensatory noise generated by daily habits, accurately captures hidden physiological and psychological decline compensations, and achieves high-precision, very early, imperceptible warnings.

Description

Technical Field

[0001] This invention relates to the field of image analysis technology, specifically to a method and system for image-based assessment of the daily activity abilities of elderly people living at home. Background Technology

[0002] With the widespread application of computer vision technology in smart elderly care, assessment methods for the daily activity abilities of the elderly based on continuous image sequences have become the mainstream monitoring approach. Existing technologies can extract the three-dimensional coordinates of human posture and combine them with a static environmental spatial model to continuously monitor the elderly's movement trajectory and contact with surrounding physical boundaries. This non-contact image assessment method can objectively record the spatial displacement characteristics and frequency of environmental interactions without interfering with the elderly's daily lives, thus providing necessary data support for assessing their basic motor functions and enabling automated tracking and preliminary screening of daily activity status in home-based elderly care scenarios.

[0003] However, existing technologies typically rely solely on superficial physical contact events or relatively simple deviations from the trajectory to determine the functional status of the assessed individual, failing to delve deeper into the true driving logic behind complex behaviors. Specifically, due to the high individual variability in the daily movement processes of the assessed individuals, existing assessment logic struggles to distinguish whether their contact behaviors with the surrounding physical environment stem from long-established unconscious personal habits or from a forced compensatory response resulting from a hidden decline in their physiological functions or an extreme fear of falling. This crude judgment mechanism, which equates all environmental contact behaviors with a decline in ability, is highly susceptible to introducing a large amount of spurious compensatory noise during monitoring, leading to frequent false alarms and causing the final assessment results to deviate significantly from the assessed individual's true physical condition. This makes it difficult to achieve high-precision, early-stage diagnosis and warning of hidden declines in daily activity abilities. Summary of the Invention

[0004] To address the problems in related technologies, this invention provides an image-based assessment method and system for the daily activity abilities of elderly people living at home, thereby overcoming the aforementioned technical problems in existing related technologies.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a method for image-based assessment of the daily activity abilities of elderly people living at home, comprising the following steps:

[0006] Acquire continuous temporal images of home scenes and perform feature analysis to construct a physical boundary model containing surface normal vectors, and extract the three-dimensional coordinates of the centroid and the three-dimensional coordinates of the upper limb end in the continuous temporal sequence.

[0007] The centroid velocity vector is calculated based on the three-dimensional coordinates of the centroid, and the upper limb velocity vector is calculated based on the three-dimensional coordinates of the upper limb distal end. At the same time, the shortest spatial distance from the three-dimensional coordinates of the upper limb distal end to the physical boundary model is calculated.

[0008] Based on the shortest spatial distance, the centroid velocity vector, the upper limb velocity vector, and the surface normal vector, the continuous temporal sequence is divided into effective supported frames and unsupported free frames;

[0009] The trajectory bias cost between the actual trajectory distance and the ideal straight line distance is calculated based on the three-dimensional coordinates of the centroid. The number of effective support frames and the trajectory bias cost are fused and calculated using a preset initial weight coefficient to obtain an initial environmental support feature value that characterizes the explicit physical compensation tendency.

[0010] Determine whether the initial environment-dependent feature value is greater than a preset trigger threshold;

[0011] If so, extract the three-dimensional coordinates of the centroid corresponding to the unsupported free frame, calculate the centroid fluctuation variance of the unsupported free frame in the preset vertical direction and map it into gait stability confidence.

[0012] The initial weight coefficients are adjusted based on the gait stability confidence, and the fusion calculation is re-executed using the adjusted weight coefficients to generate corrected environmental dependency feature values.

[0013] The corrected environmental feature value is compared with a preset historical baseline to obtain the variation deviation. If the variation deviation is greater than a preset variation threshold, a decline warning signal is generated.

[0014] Preferably, dividing the continuous temporal sequence into supported frames and unsupported free frames based on the shortest spatial distance, the centroid velocity vector, the upper limb velocity vector, and the surface normal vector includes the following steps:

[0015] Determine whether the shortest spatial distance is less than or equal to a preset distance threshold;

[0016] If not, the current frame is classified as an unattached, free-floating frame;

[0017] If so, and it satisfies any of the following conditions:

[0018] If the magnitude of the centroid velocity vector is less than or equal to a preset static velocity threshold, or if the centroid velocity vector, the upper limb velocity vector, and the surface normal vector satisfy a preset dynamic cooperative sliding condition, then the current frame is classified as a valid support frame.

[0019] The centroid velocity vector, the upper limb velocity vector, and the surface normal vector satisfy the preset dynamic cooperative sliding condition, including the following steps:

[0020] Based on the upper limb velocity vector and the surface normal vector, calculate the tangential velocity component of the upper limb velocity vector on the surface of the physical boundary model;

[0021] Calculate the ratio of the magnitude of the tangential velocity component to the magnitude of the upper limb velocity vector, and calculate the spatial angle between the upper limb velocity vector and the center of mass velocity vector;

[0022] When the ratio of tangential velocities is greater than a preset tangential ratio threshold and the spatial angle is less than a preset angle threshold, it is determined that the upper limb velocity vector and the surface normal vector satisfy the dynamic cooperative sliding condition.

[0023] Preferably, the calculation of the trajectory bias cost between the actual trajectory distance and the ideal straight-line distance based on the three-dimensional coordinates of the centroid includes the following steps:

[0024] Obtain the first centroid three-dimensional coordinates corresponding to the start frame and the second centroid three-dimensional coordinates corresponding to the end frame of the continuous time sequence from the centroid three-dimensional coordinates, and calculate the ideal straight-line distance between the first centroid three-dimensional coordinates and the second centroid three-dimensional coordinates;

[0025] The centroid three-dimensional coordinates of adjacent frames in the continuous time sequence are traversed, and the coordinate displacement between adjacent frames is accumulated to obtain the actual trajectory distance;

[0026] The difference between the actual trajectory distance and the ideal straight line distance is calculated, and the ratio of the difference to the ideal straight line distance is determined as the trajectory bias cost.

[0027] The process of traversing the three-dimensional coordinates of the centroid of adjacent frames in the continuous time sequence and accumulating the coordinate displacement between adjacent frames to obtain the actual trajectory distance includes the following steps:

[0028] Extract the two-dimensional projection coordinates of the centroid three-dimensional coordinates on the horizontal plane corresponding to each of the adjacent frames;

[0029] The two-dimensional projected coordinates are denoised using a preset smoothing filtering algorithm to obtain smoothed two-dimensional projected coordinates.

[0030] Calculate and sum the Euclidean distances between adjacent smoothed 2D projected coordinates, and determine the summed result as the actual trajectory distance.

[0031] Preferably, obtaining the initial environmental dependency characteristic value characterizing the dominant physical compensation tendency includes the following steps:

[0032] Obtain the total number of frames in the continuous time sequence, calculate the ratio of the number of effective dependent frames to the total number of frames, and obtain the normalized dependent frame rate;

[0033] Extract the preset first initial weight coefficient and second initial weight coefficient;

[0034] The first feature term is obtained by multiplying the dependent frame rate by the first initial weight coefficient, and the second feature term is obtained by multiplying the trajectory bias cost by the second initial weight coefficient.

[0035] The first feature term and the second feature term are summed to obtain the initial environmental dependency feature value.

[0036] Preferably, calculating the centroid fluctuation variance of the unsupported free frame in a preset vertical direction and mapping it to gait stability confidence includes the following steps:

[0037] Extract the vertical coordinate components of the centroid three-dimensional coordinates in the preset vertical direction corresponding to each of the aforementioned unsupported free frames;

[0038] Calculate the mean of all the vertical coordinate components, and calculate the variance of all the vertical coordinate components based on the mean to obtain the centroid fluctuation variance;

[0039] Calculate the ratio of the preset standard fluctuation variance to the centroid fluctuation variance, and generate a gait stability confidence score based on the ratio using a preset mapping function; wherein the gait stability confidence score is negatively correlated with the centroid fluctuation variance.

[0040] Preferably, generating the corrected environment-dependent feature value includes the following steps:

[0041] Extract all the valid support frames in the continuous time series, and identify all the continuous support frame segments in chronological order;

[0042] Obtain the frame length of each consecutive supporting frame segment, filter out the segment with the longest frame length, and calculate the ratio of the longest frame length of the segment to the number of effective supporting frames to obtain the supporting time continuity index.

[0043] The gait stability confidence and the time continuity index are input into a preset nonlinear mapping function;

[0044] When the gait stability confidence is less than or equal to the preset confidence threshold, a positive gain adjustment factor greater than 1 is output.

[0045] When the gait stability confidence is greater than the preset confidence threshold and the reliance time continuity index is less than the preset continuity threshold, a negative decay adjustment factor less than 1 is output.

[0046] When the gait stability confidence is greater than the preset confidence threshold and the reliance time continuity index is greater than or equal to the preset continuity threshold, a non-monotonic inversion mechanism is triggered, and a higher-order gain adjustment factor greater than 1 is output in the reverse direction.

[0047] The adjusted first weight coefficient is obtained by multiplying the output adjustment factor by the first initial weight coefficient.

[0048] The adjusted first weight coefficient is used to perform reverse compensation on the second initial weight coefficient to obtain the adjusted second weight coefficient.

[0049] Using the adjusted first weight coefficient and the adjusted second weight coefficient, the dependent frame rate and the trajectory bias cost are summed again to generate the corrected environment dependent feature value.

[0050] The present invention also includes an image assessment system for the daily activity abilities of elderly people living at home, comprising:

[0051] The data acquisition module is used to acquire continuous temporal images of home scenes and perform feature analysis to construct a physical boundary model containing surface normal vectors, and extract the three-dimensional coordinates of the centroid and the three-dimensional coordinates of the upper limb end of the human body pose in the continuous temporal sequence.

[0052] The kinematics calculation module is used to calculate the centroid velocity vector based on the three-dimensional coordinates of the centroid and the upper limb velocity vector based on the three-dimensional coordinates of the upper limb distal end, and simultaneously calculate the shortest spatial distance from the three-dimensional coordinates of the upper limb distal end to the physical boundary model;

[0053] The temporal sequence division module is used to divide the continuous temporal sequence into effective supported frames and unsupported free frames based on the shortest spatial distance, the centroid velocity vector, the upper limb velocity vector, and the surface normal vector.

[0054] The initial feature generation module is used to calculate the trajectory bias cost between the actual trajectory distance and the ideal straight line distance based on the three-dimensional coordinates of the centroid, and to perform a fusion calculation on the number of effective support frames and the trajectory bias cost using a preset initial weight coefficient to obtain an initial environmental support feature value that characterizes the explicit physical compensation tendency.

[0055] The feedback adjustment module is used to determine whether the initial environmental support feature value is greater than a preset trigger threshold; if so, it extracts the centroid three-dimensional coordinates corresponding to the unsupported free frame, calculates the centroid fluctuation variance of the unsupported free frame in a preset vertical direction and maps it to gait stability confidence; it adjusts the initial weight coefficients based on the gait stability confidence, and re-executes the fusion calculation through the adjusted weight coefficients to generate the corrected environmental support feature value;

[0056] An anomaly warning module is used to compare the corrected environmental feature values ​​with a preset historical baseline to obtain the variation deviation. If the variation deviation is greater than a preset variation threshold, a degradation warning signal is generated.

[0057] By employing the above technical solution, the present invention provides an image-based assessment method and system for the daily activity abilities of elderly people living at home, which has at least the following beneficial effects:

[0058] 1. This invention extracts multi-dimensional kinematic features from continuous temporal images of home scenes, and sequentially completes physical boundary modeling, dual-state frame classification, initial feature fusion, gait perception-driven adaptive weight adjustment, and individualized historical baseline comparison. It constructs a complete evaluation closed loop from original image perception to decline warning signal output, which can continuously, objectively, and automatically quantify and track the longitudinal decline of the elderly’s environmental compensatory behavior in real daily activity scenarios.

[0059] 2. This invention introduces a dynamic cooperative sliding condition based on a three-dimensional joint judgment of tangential velocity components, tangential velocity ratio, and the spatial angle between upper and lower limb velocities. From a kinematic perspective, it accurately identifies functional contact in each frame of a continuous time sequence. It accurately distinguishes two types of contact behaviors—sliding along a surface for leverage and purposeful operations in the vertical direction—as effective support frames and unsupported free frames. This effectively avoids interference from non-leveraging interactive actions such as picking up objects and pressing switches on the statistics of support frames, and improves the data accuracy of the initial environmental support feature value calculation.

[0060] 3. This invention quantifies the overt physical compensation tendency from two orthogonal dimensions: space and time. It obtains an initial environmental dependence feature value by weighted fusion of trajectory bias cost and dependence frame rate, and uses a preset trigger threshold as the initiation condition for deep diagnosis. Trajectory bias cost captures the degree of detour in the elderly's walking path caused by dependence behavior, while dependence frame rate reflects the time proportion of dependence behavior in the entire time series. The complementary fusion of these two factors allows for effective differentiation between simple path detours and simple frequent but brief contact. Simultaneously, the conditional trigger mechanism ensures that the subsequent complete adaptive adjustment process is initiated only for cases where the initial feature value exceeds the threshold, reducing the overall computational load of the system while ensuring assessment accuracy.

[0061] 4. This invention extracts the vertical fluctuation variance of the centroid of unsupported free frames and maps it to gait stability confidence. Combined with the support time continuity index, a three-branch nonlinear mapping function is designed to adaptively adjust the initial weight coefficients: a positive gain is applied to those with truly unstable gait, a negative attenuation is applied to those with habitual light touches, and a non-monotonic inversion higher-order gain is triggered for those with persistently high support due to extreme fear of falling. This enables the system to adaptively distinguish different clinical scenarios, reduce the false positive warning rate, and improve the clinical credibility of the assessment results. Attached Figure Description

[0062] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0063] Figure 1 A flowchart of the image assessment method for daily activity ability of elderly people living at home provided by the present invention;

[0064] Figure 2 This is a schematic diagram of the modules of the home-based elderly care daily activity ability image assessment system provided by the present invention. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] Exemplary method:

[0067] In home-based elderly care settings, the decline in daily activity abilities among the elderly is often not overt. When the elderly experience a decrease in walking stability, two types of compensatory strategies emerge in their behavior: one is overt physical compensation, which involves using the physical environment such as walls, furniture edges, and handrails for leverage; the other is covert psychological compensation, which manifests as habits such as taking detours and slowing down due to fear of falling. These two types of compensation are difficult to identify through daily observation or subjective scales, but changes in their frequency and intensity are important early signs of covert decline in motor function in the elderly. Existing methods for assessing home-based activity abilities mostly rely on periodic in-home testing or fixed-movement scale scoring, which cannot continuously and objectively quantify and monitor compensatory behaviors in the elderly's actual daily activities, and suffer from limitations such as sparse sampling and subjective bias.

[0068] To address the aforementioned issues, this application proposes an image-based assessment method for the daily activity abilities of elderly people living at home, such as... Figure 1 As shown, the method includes the following steps:

[0069] S1. Acquire continuous temporal images of the home scene and perform feature analysis to construct a physical boundary model containing surface normal vectors, and extract the three-dimensional coordinates of the centroid and the three-dimensional coordinates of the upper limb end in the continuous temporal sequence.

[0070] It should be noted that this step is the data acquisition and scene modeling stage, which forms the data foundation of the entire evaluation system. The acquisition end can deploy an RGB-D depth camera (such as Intel RealSense, Microsoft Azure Kinect, etc.) or a multi-view vision system to obtain a temporal image sequence containing color and depth information through continuous shooting. The system performs feature analysis on the acquired images to identify potential support surfaces in the home environment, such as sofa armrests, bed edges, table edges, and walls, and extracts unit surface normal vectors for each support surface to construct a physical boundary model. The normal vector extraction uses a local point cloud plane fitting method, with the normal vector direction indicating the orientation of the support surface, providing a geometric basis for subsequent contact determination. Meanwhile, the system extracts human pose in continuous time sequence based on human pose estimation algorithms (such as MediaPipe Holistic, OpenPose, etc.) to obtain the three-dimensional coordinates of the centroid and the three-dimensional coordinates of the upper limb distal end. The three-dimensional coordinates of the centroid are taken from the coordinates of the midpoint of the hip joint in three-dimensional space, and the three-dimensional coordinates of the upper limb distal end are taken from the coordinates of the wrist joint or the key point of the hand distal end. The above coordinates are extracted frame by frame in continuous time sequence to form the time sequence of the three-dimensional coordinates of the centroid and the time sequence of the three-dimensional coordinates of the upper limb distal end, providing raw data for subsequent kinematic feature calculation.

[0071] S2. Calculate the centroid velocity vector based on the three-dimensional coordinates of the centroid, and calculate the upper limb velocity vector based on the three-dimensional coordinates of the upper limb end, while simultaneously calculating the shortest spatial distance from the three-dimensional coordinates of the upper limb end to the physical boundary model;

[0072] Specifically, this step obtains velocity vectors and spatial distance information representing motion state by performing kinematic differentiation and distance query on the three-dimensional coordinate sequence extracted in the previous step, providing quantitative criteria for subsequent frame classification.

[0073] The centroid velocity vector and the upper limb velocity vector are obtained by dividing the difference between adjacent frame coordinates by the inter-frame time interval, respectively reflecting the overall body motion state and the upper limb motion direction; the inter-frame time interval is determined by the acquisition frequency (taking an acquisition frequency of 30fps as an example, the inter-frame time interval = 1 / 30 ≈ 0.033 seconds).

[0074] The shortest spatial distance represents the shortest Euclidean distance from the end of the upper limb in the current frame to all supporting surfaces in the physical boundary model. It is obtained by searching for the nearest point of each supporting surface in the model and is used to quantify the proximity between the upper limb and the supporting surface.

[0075] S3. Based on the shortest spatial distance, the centroid velocity vector, the upper limb velocity vector, and the surface normal vector, the continuous temporal sequence is divided into effective supported frames and unsupported free frames;

[0076] It should be noted that this step, by performing multi-condition joint judgment on the multi-dimensional kinematic features obtained in the previous step, accurately marks each frame in the continuous time sequence as either a valid support frame or a free-floating frame, directly determining the input for subsequent initial feature value calculations. This dual-state classification mechanism not only distinguishes whether the elderly person has physical contact with the environment, but also further identifies the functional nature of the contact (support or vertical operation), effectively avoiding misclassification of vertical hand interactions such as picking up a cup or pressing a switch as environmental support behavior.

[0077] The specific steps for dividing the continuous temporal sequence into effective supported frames and unsupported free frames based on the shortest spatial distance, the centroid velocity vector, the upper limb velocity vector, and the surface normal vector are as follows:

[0078] S31. Determine whether the shortest spatial distance is less than or equal to a preset distance threshold;

[0079] Specifically, the shortest spatial distance is the primary condition for determining whether the upper limb end is close to or in contact with a physical boundary. Its function is to "gated the distance" for subsequent fine kinematic judgments, excluding the detached state of the elderly when they are far from any environmental boundary. The preset distance threshold can be set according to the sensor accuracy and experimental calibration results, typically ranging from 0.05m to 0.15m. This range is mainly based on two engineering and physiological considerations: First, mainstream depth sensors (such as RGB-D cameras) have approximately 0.01m to 0.03m of depth ranging system noise at a typical indoor detection distance of 2 to 4 meters; second, human skeletal key point detection algorithms typically extract the "wrist joint center" or "palm center," which has a physiological thickness difference of approximately 0.03m to 0.05m from the physical boundary where the hand skin actually contacts the wall, and may be affected by the thickness of the elderly's clothing. Therefore, if the threshold is less than 0.05m, it is very easy to miss the detection due to physiological thickness and sensor noise (judging real contact as free contact); if the threshold is greater than 0.15m, it is easy to misjudge the normal swinging of the arm towards the wall as contact.

[0080] S32. If not, then divide the current frame into an unsupported free frame;

[0081] In this embodiment, if the shortest spatial distance of a certain frame is 0.45m (the elderly person is walking in the center of the room), then the frame is directly marked as an unsupported free frame, and its centroid coordinates are added to the free frame set for variance statistics in subsequent gait stability confidence calculation.

[0082] S33. If so, and if any of the following conditions are met: the magnitude of the centroid velocity vector is less than or equal to a preset static velocity threshold, or the centroid velocity vector, the upper limb velocity vector, and the surface normal vector satisfy a preset dynamic cooperative sliding condition, then the current frame is classified as a valid support frame.

[0083] It should be noted that this step designs two parallel effective support determination paths, corresponding to two typical support behavior patterns of the elderly: static support and dynamic sliding leverage. The static support path is identified by the magnitude of the centroid velocity vector, corresponding to scenarios such as the elderly resting against a wall or standing while holding onto a table; the dynamic sliding leverage path is identified by dynamic coordinated sliding conditions, corresponding to scenarios where the elderly slide along a wall while simultaneously walking. The OR relationship between the two paths ensures that both types of support behaviors can be accurately captured.

[0084] The preset static velocity threshold is typically set between 0.02 m / s and 0.05 m / s. This range is a trade-off between human biomechanical characteristics and preventing the underestimation of extremely slow gait: First, when the human body is naturally upright or resting against a wall, it physically approximates an inverted pendulum system. To maintain postural balance, physiological swaying of the center of gravity is inevitable. This, combined with the coordinate jitter noise of the attitude estimation algorithm between consecutive frames, means that an absolute 0 m / s does not exist in reality. If the threshold is below 0.02 m / s, this normal physiological micro-swaying and system noise will cause resting in place to be frequently misjudged as movement. Conversely, if the threshold is set too high (e.g., greater than 0.05 m / s), the extremely slow, shuffling movements characteristic of the elderly or frail can easily be incorrectly classified as standing rest, thus masking the true gait characteristics. Therefore, this extremely narrow range of 0.02 m / s to 0.05 m / s is an engineering critical value that simultaneously accommodates both physiological and systemic micro-swaying and keenly captures extremely slow gait.

[0085] Furthermore, the process of the centroid velocity vector, the upper limb velocity vector, and the surface normal vector satisfying the preset dynamic cooperative sliding condition includes the following steps:

[0086] First, based on the upper limb velocity vector and the surface normal vector, calculate the tangential velocity component of the upper limb velocity vector on the surface of the physical boundary model;

[0087] Specifically, the formula for calculating the tangential velocity component is as follows:

[0088] ,

[0089] in, The unit normal vector at the nearest boundary point is provided by the physical boundary model; express Projection along the normal vector direction, This represents the upper limb velocity vector. Indicates the tangential velocity component;

[0090] Next, the ratio of the magnitude of the tangential velocity component to the magnitude of the upper limb velocity vector is calculated, and the spatial angle between the upper limb velocity vector and the center of mass velocity vector is calculated; the tangential velocity ratio... Angle with space The calculation formula is:

[0091] ,

[0092] ,

[0093] in, Represents the velocity vector of the center of mass;

[0094] The tangential velocity ratio measures the proportion of sliding along the wall during upper limb movement (0-1). The closer the value is to 1, the closer the hand movement is to the tangential direction of the wall. The spatial angle measures the degree of directional coordination between the upper limb velocity and the center of mass velocity. The smaller the value, the more consistent the direction of hand movement is with the overall direction of body movement.

[0095] Finally, when the tangential velocity ratio is greater than a preset tangential ratio threshold and the spatial angle is less than a preset angle threshold, it is determined that the upper limb velocity vector and the surface normal vector satisfy the dynamic cooperative sliding condition.

[0096] Among them, the preset tangential ratio threshold represents the minimum tangential motion ratio required for the upper limb to effectively slide in contact with the ground; the preset angle threshold represents the maximum tolerable biomechanical angle between the upper limb sliding trajectory and the trunk's direction of travel when the human body is moving forward in coordination.

[0097] Preferably, the preset tangential ratio threshold is typically set between 0.70 and 0.85. This preset tangential ratio threshold is based on the principles of surface contact physics, requiring that the vast majority (over 70%) of the absolute motion components of the upper limb distals must be distributed within a tangential plane parallel to the physical boundary. This constraint can extremely accurately filter out purposeful interactive actions with significant normal (vertical) forces in the daily lives of the elderly, such as pushing or pulling doors and windows, pressing wall switches, or reaching for objects directly in front of them—actions requiring no leverage. The preset angle threshold is typically set between 30° and 60°. The setting of this angle range fully considers the biomechanical characteristics of human walking: In normal alternating gait, in order to maintain trunk balance, the human upper limb will produce a natural compensatory arm swing, which inevitably results in a certain physiological angle between the instantaneous velocity vector of the hand and the overall direction of the center of mass. Therefore, it is impossible to demand that the two be absolutely parallel from an engineering perspective. However, if this spatial angle is too large (for example, exceeding 60°), it indicates from a kinematic perspective that the sliding trajectory of the hand deviates significantly from the direction of the body's center of gravity (such as wiping the table laterally, turning to the side and back to feel around, etc.), violating the cooperative physical law of using momentum to move forward. Only when the above dual constraints are satisfied simultaneously can the system reliably determine that the elderly person is using the boundary surface for dynamic gliding and momentum.

[0098] S4. Calculate the trajectory bias cost between the actual trajectory distance and the ideal straight line distance based on the three-dimensional coordinates of the centroid. Use the preset initial weight coefficient to fuse the number of effective support frames and the trajectory bias cost to obtain the initial environmental support feature value that characterizes the explicit physical compensation tendency.

[0099] It should be noted that this step quantifies the overt physical compensation tendency from two orthogonal perspectives: spatial and temporal dimensions. A single feature value is obtained through weighted fusion, providing a unified evaluation metric for subsequent trigger threshold judgment. Specifically, trajectory bias cost captures the degree of detour in the elderly's walking path caused by reliance behavior, while reliance frame rate reflects the proportion of time spent on reliance behavior throughout the entire time series. The complementarity of these two metrics allows for the differentiation between simple path detours (such as detours caused by room layout) and simple frequent, brief contact (such as habitual light touches), avoiding misjudgments based on single-dimensional indicators.

[0100] The specific steps for obtaining the initial environmental dependence characteristic values ​​that characterize the overt physical compensation tendency are as follows:

[0101] S41. Obtain the first centroid three-dimensional coordinates corresponding to the start frame and the second centroid three-dimensional coordinates corresponding to the end frame of the continuous time sequence from the centroid three-dimensional coordinates, and calculate the ideal straight-line distance between the first centroid three-dimensional coordinates and the second centroid three-dimensional coordinates.

[0102] The ideal straight-line distance is defined as the Euclidean distance between the three-dimensional coordinates of the centroids of the first and last frames in a continuous time series, representing the theoretical length of the shortest possible path for the elderly person within that time series; the calculation formula is as follows:

[0103] ,

[0104] in, The total number of frames in continuous time sequence. and These are the three-dimensional coordinates of the centroids of the start and end frames, respectively.

[0105] S42. Traverse the three-dimensional coordinates of the centroid of adjacent frames in the continuous time sequence, accumulate the coordinate displacement between adjacent frames, and obtain the actual trajectory distance;

[0106] Furthermore, the step of traversing the three-dimensional coordinates of the centroid of adjacent frames in the continuous time sequence, and accumulating the coordinate displacement between adjacent frames to obtain the actual trajectory distance includes the following steps:

[0107] The first step is to extract the two-dimensional projection coordinates of the three-dimensional centroid coordinates of each adjacent frame onto the horizontal plane; the projection on the horizontal plane eliminates the normal gait fluctuation component in the vertical direction (Z-axis), so that the trajectory bias cost focuses on reflecting the Z-shaped sway on the horizontal plane, avoiding interference from the vertical fluctuation of normal walking on the trajectory length calculation.

[0108] The second step is to use a preset smoothing filtering algorithm to denoise the two-dimensional projection coordinates to obtain smoothed two-dimensional projection coordinates.

[0109] The preset smoothing filtering algorithm can employ Savitzky-Golay filtering, mean filtering, or median filtering. The purpose of smoothing filtering is to remove high-frequency coordinate fluctuations introduced by camera detection noise or short-term jitter, while preserving the true bias characteristics of the walking trajectory.

[0110] The third step is to calculate and sum the Euclidean distances between adjacent smoothed 2D projected coordinates, and then determine the summed result as the actual trajectory distance. The calculation formula is as follows:

[0111] ,

[0112] in, This represents the actual trajectory distance. and They represent the first Frame and the The smoothed two-dimensional projected coordinates of the frame;

[0113] S43. Calculate the difference between the actual trajectory distance and the ideal straight line distance, and determine the ratio of the difference to the ideal straight line distance as the trajectory bias cost;

[0114] Specifically, trajectory bias cost quantifies the degree of deviation of the elderly person's actual walking path from the ideal straight path; the calculation formula is as follows:

[0115] ,

[0116] in, Indicates the cost of trajectory bias, if This indicates that the elderly person moves along a completely straight path. The larger the value, the more circuitous the path. Older adults with unstable gait often exhibit larger values. The value reflects its wall-feeling bypass mode when it relies on environmental supports.

[0117] S44. Obtain the total number of frames in the continuous time sequence, calculate the ratio of the number of effective support frames to the total number of frames, and obtain the normalized support frame rate; wherein, the larger the value of the support frame rate, the higher the proportion of time the elderly rely on the environment in this time sequence.

[0118] S45. Extract the preset first initial weight coefficient and the preset second initial weight coefficient; preset initial weight coefficient Set during system initialization; can be set to... =0.6, =0.4, and the sum of the two is 1, which reflects the relative contribution of the cost of relying on frame rate and trajectory bias in the initial evaluation.

[0119] S46. Multiply the dependent frame rate by the first initial weight coefficient to obtain a first feature term, and multiply the trajectory bias cost by the second initial weight coefficient to obtain a second feature term;

[0120] S47. Summing the first feature term and the second feature term yields the initial environmental dependency feature value. The calculation formula is:

[0121] ,

[0122] in, This indicates that it is based on the frame rate.

[0123] Specifically, the initial environmental dependency feature value output in this step will be compared with a preset trigger threshold in step S5. If the initial environmental dependency feature value is greater than the preset trigger threshold, the system enters the deep diagnostic process based on gait stability confidence; otherwise, the initial environmental dependency feature value can be directly used as the final evaluation value and input into step S7 for historical baseline comparison. This conditional triggering design effectively saves computational resources, initiating the complete adaptive adjustment process only for cases that truly require in-depth analysis.

[0124] S5. Determine whether the initial environment-dependent feature value is greater than the preset trigger threshold; if so, extract the centroid three-dimensional coordinates corresponding to the unsupported free frame, calculate the centroid fluctuation variance of the unsupported free frame in the preset vertical direction and map it to gait stability confidence.

[0125] It should be noted that in this step, when the initial environmental support feature value exceeds the trigger threshold, the system extracts the centroid vertical coordinate sequence of the unsupported free frames marked in step S3, calculates its fluctuation variance, and maps the fluctuation variance to gait stability confidence. Gait stability confidence represents a quantitative judgment of whether the current support behavior is driven by true gait instability, and is used to adaptively adjust the driving weights in step S6. The purpose of this step is to utilize only the centroid vertical fluctuation data during the elderly person's natural walking (unsupported frames), rather than the data from the support behavior itself, thereby eliminating vertical displacement interference caused by active leaning and ensuring the objectivity of gait stability assessment.

[0126] Specifically, the preset trigger threshold is typically set between 0.20 and 0.30. This range is based on the baseline statistical distribution of daily natural activities of healthy elderly individuals. In engineering practice, the mean of the historical distribution of the initial environmental characteristics of the healthy control group plus 1 to 1.5 times the standard deviation is usually taken as the calibration benchmark. This setting balances the sensitivity of the system's early warning and the underlying computing power: if the threshold is set too low (e.g., below 0.15), it is easy to over-amplify the extremely occasional wall-leaning behavior of healthy elderly people in their normal lives (such as occasional leaning for rest or lightly touching to find a sense of space), resulting in a large influx of normal samples into the subsequent complex free frame variance extraction and nonlinear mapping process, causing a "computing power explosion" and increasing the false positive rate; conversely, if the threshold is set too high (e.g., above 0.35), it will cause the system to be too slow in its judgment, and it is easy to miss hidden high-risk signals in the very early stages of functional decline when the compensatory characteristics are not yet extremely significant.

[0127] The specific steps for calculating the centroid fluctuation variance of the unsupported free frame in the preset vertical direction and mapping it to gait stability confidence are as follows:

[0128] S51. Extract the vertical coordinate components of the centroid three-dimensional coordinates of each of the aforementioned unsupported free frames in a preset vertical direction; the preset vertical direction is usually the Z-axis or Y-axis of the world coordinate system (depending on the camera installation method), and denote the i-th frame in the set of unsupported free frames as the vertical coordinate component of the centroid three-dimensional coordinates in a preset vertical direction. The vertical coordinate component corresponding to the frame is ,common A free-floating frame without any support.

[0129] S52. Calculate the mean of all the vertical coordinate components, and calculate the variance of all the vertical coordinate components based on the mean to obtain the centroid fluctuation variance; vertical coordinate mean With centroid fluctuation variance The calculation formula is:

[0130] ,

[0131] ,

[0132] Specifically, the centroid fluctuation variance reflects the degree of dispersion of the centroid fluctuation in the vertical direction when the elderly walk freely; a healthy gait has regular bimodal vertical centroid fluctuations (step length corresponds to a complete sine wave). The gait is moderate; however, older adults with unstable gait exhibit irregular vertical undulations. High levels suggest insufficient lower limb muscle strength or decreased balance.

[0133] S53. Calculate the ratio of the preset standard fluctuation variance to the centroid fluctuation variance, and generate a gait stability confidence score based on the ratio using a preset mapping function; wherein, the gait stability confidence score is negatively correlated with the centroid fluctuation variance; the mapping function is as follows:

[0134] ,

[0135] in, The standard deviation is the pre-defined standard deviation (obtained from gait data of healthy elderly people, with a typical value of 0.0015 m²). For monotonically increasing mapping functions (such as the Sigmoid function or piecewise linear functions). When the ratio Smaller than average Too low; when When it is close to the standard value, Approaching 1.0. ,in, A lower gait indicates poor gait stability, while a higher gait indicates a gait that is close to normal. This negative correlation makes... It can directly quantify gait stability.

[0136] For example, in this implementation example, =120 frames (unsupported free-floating frames), calculated as follows , .Pick ,but After Sigmoid mapping, It is significantly lower than the preset reliability threshold. (e.g., 0.50). This indicates that older adults have poor gait stability when walking freely, further confirming the high [stability / stability] in step S4. The value has a real physiological decline meaning, rather than simply a habitual dependence, triggering the positive gain adjustment of step S6.

[0137] S6. Adjust the initial weight coefficients based on the gait stability confidence, and re-execute the fusion calculation using the adjusted weight coefficients to generate the corrected environmental dependency feature values;

[0138] The specific steps for generating the corrected environment-dependent feature values ​​are as follows:

[0139] S61. Extract all the effective support frames in the continuous time sequence, and identify all the continuous support frame segments in chronological order; wherein, the continuous support frame segment represents a sequence of frames that are temporally adjacent and are all marked as effective support frames, reflecting the continuous time period during which the elderly person continuously relies on a certain support.

[0140] S62. Obtain the frame length of each consecutive supporting frame segment, filter out the segment with the longest frame length, and calculate the ratio of the longest frame length to the number of valid supporting frames to obtain the supporting time continuity index; Supporting time continuity index The calculation formula is:

[0141] ,

[0142] in, The maximum frame length among all consecutive dependent frame segments; To effectively rely on the total number of frames; A higher value indicates that the reliance behavior tends to be more concentrated and continuous (such as holding onto the same support for a long time), which is especially common in elderly people who habitually walk by supporting themselves against a wall or have an extreme fear of falling; a lower value indicates that the reliance is scattered and intermittent (small and frequent contact), which is more in line with the real pathological behavior pattern of frequently seeking support due to gait instability.

[0143] S63. Input the gait stability confidence and the reliance time continuity index into a preset nonlinear mapping function; when the gait stability confidence is less than or equal to a preset confidence threshold, output a positive gain adjustment factor greater than 1; when the gait stability confidence is greater than the preset confidence threshold and the reliance time continuity index is less than a preset continuity threshold, output a negative attenuation adjustment factor less than 1; when the gait stability confidence is greater than the preset confidence threshold and the reliance time continuity index is greater than or equal to the preset continuity threshold, trigger a non-monotonic inversion mechanism and output a higher-order gain adjustment factor greater than 1 in the reverse direction; the nonlinear mapping function is as follows:

[0144] ,

[0145] in, This indicates a preset continuity threshold; , as well as These represent the positive gain base coefficient, the negative attenuation base coefficient, and the inverted higher-order gain coefficient, respectively. , as well as These represent the nonlinear adjustment index, the continuity sensitivity index, and the exponential surge factor, respectively.

[0146] The preset continuity threshold is typically set between 0.70 and 0.85. This value is considered the golden ratio for distinguishing between "habitual multi-segment light touching" and "psychological clinging to the wall." In clinical observation, if a single continuous wall-touching frame accounts for more than 70% of the total wall-touching frames, it indicates that the subject's reliance behavior exhibits strong temporal clustering, signifying a severe lack of confidence in independent walking and a fear of falling; conversely, it suggests a greater tendency towards occasional habitual touching.

[0147] Specifically, the positive gain base coefficient, the negative attenuation base coefficient, and the inverted higher-order gain coefficient determine the absolute base amplitude of the weight adjustment.

[0148] Preferred positive gain base coefficient The value is typically set between 0.2 and 0.4; the rationale is that when purely physiological gait instability is detected, a moderate gain of up to 1.2 to 1.4 times is applied to the environmental dependence feature to amplify the true compensatory signal, but the value should not be too large to avoid collapsing the assessment system. Negative attenuation baseline coefficient. The value is typically set between 0.3 and 0.5; the rationale is that for confirmed habitual taps (false compensation), the system needs to reduce their weight to 50% to 70% of the original. This lower limit ensures that the interference of false compensation is reduced below the trigger threshold, effectively curbing false positives. (Inverting higher-order gain coefficients) The value is typically between 0.4 and 0.6; this is the base penalty for triggering the "fear of falling" reversal mechanism, and its value must be greater than [amount missing]. Because covert psychological compensation is more difficult to detect and has a higher risk factor than overt physiological instability, the system needs to directly impose a higher-level penalty threshold of 1.4 to 1.6 times or more.

[0149] Specifically, the nonlinear adjustment index, the continuity sensitivity index, and the exponential surge factor determine the curvature of the three-dimensional mapped surface and its sensitivity to change.

[0150] Preferably, the nonlinear adjustment index Typically, the value is 2.0 (squared) or greater; its physical meaning is that the lower the gait confidence (the more unstable), the increase in weight is not uniform, but rather accelerates parabolically, imposing a severe exponential amplification on severely unstable subjects. Continuity Sensitivity Index Typically set to 1.0 or 2.0, this value controls the smoothness of the decay of habitual light-touch features, ensuring that the weight decay converges smoothly when transitioning from short wall-touching to long wall-touching, preventing abrupt step oscillations in the algorithm output. Exponential Surge Factor The value is usually set between 3.0 and 5.0. As the core driving force of the third branch, its value is set relatively high. It is intended to trigger a violent burst of the exponential function when the continuity index breaks through the threshold. Specifically, once the elderly person supports the wall for a longer period of time than the safety line, even if it is only for one second longer, the adjustment factor will increase sharply with the increase of the continuity index, and extremely sensitively capture the psychological state of extreme fear of falling.

[0151] S64. Multiply the output adjustment factor by the first initial weight coefficient to obtain the adjusted first weight coefficient;

[0152] S65. The second initial weight coefficient is reverse-compensated using the adjusted first weight coefficient to obtain the adjusted second weight coefficient; the reverse compensation mechanism ensures that the sum of the two weight coefficients maintains the normalization constraint, that is: ,in, This represents the adjusted second weighting coefficient. This represents the first weighting coefficient after adjustment;

[0153] S66. Using the adjusted first weight coefficient and the adjusted second weight coefficient, the dependent frame rate and the trajectory bias cost are summed again to generate the corrected environment dependent feature value.

[0154] S7. Compare the corrected environmental feature value with the preset historical baseline to obtain the variation deviation. If the variation deviation is greater than the preset variation threshold, generate a decline warning signal.

[0155] Specifically, this step involves standardizing and comparing the current assessment value with the individual's historical baseline to generate a variation bias with individual adaptation, and then triggering a decline warning based on this bias. Because there are significant individual differences in the level of normal support among different elderly individuals, using a uniform absolute threshold for warnings would result in a large number of false alarms or false negatives. Individualized historical baseline comparisons, however, can capture trends relative to the individual, thus achieving more accurate longitudinal decline tracking.

[0156] The preset historical baseline is obtained through statistical analysis of the elderly person's own historical assessment data. After each assessment, the system automatically stores the generated corrected environmental dependency feature value into the elderly person's personal historical record database. The historical record database maintains recent data in a sliding window manner. Each valid assessment record corresponds to approximately one month of daily assessment data, and the historical mean and historical standard deviation are updated in real time as individual baseline parameters. For newly enrolled elderly individuals, the system... During this assessment, a pre-set group reference baseline (obtained from assessment data of healthy elderly individuals of the same age and gender) was used as a provisional baseline, pending the accumulation of individual historical records. The system will then automatically switch to an individualized baseline to ensure that it can still provide an effective initial assessment during the data accumulation period.

[0157] Preferred, A typical value for the value is 30. According to the central limit theorem in statistics, when the sample size is not less than 30, the sample mean approaches a normal distribution. The variance obtained based on this data has reliable statistical inference significance and can ensure the effectiveness of the warning threshold. At the same time, 30 assessments correspond to a daily recording cycle of about one month. While ensuring the statistical stability of the baseline, it also takes into account the baseline's ability to timely represent the recent functional status, avoiding the problem of underreporting caused by excessively long windows leading to a continuous increase in the baseline mean due to early decline data.

[0158] Specifically, the corrected environmental characteristic value is compared with a preset historical baseline to obtain the variation deviation. If the variation deviation is greater than a preset variation threshold, a degradation early warning signal is generated, including:

[0159] S71. Extract the mean and standard deviation of the preset historical baseline corresponding to the evaluated object from the historical record database, and calculate the variance deviation of the evaluated object based on the mean and standard deviation; the formula for calculating the variance deviation is as follows:

[0160] ,

[0161] in, Indicates variation bias. This represents the corrected environmental dependency characteristic value; This indicates that the current assessment value is higher than the historical average.

[0162] S72. If the variation deviation is greater than the preset variation threshold, it indicates that the current environmental characteristic value has a statistically significant abnormal increase, and the system generates a decline warning signal; otherwise, the system records the current evaluation result to the historical record database, and no warning is required.

[0163] Specifically, the preset variance threshold is typically set to 2.0; this value is strictly set according to the two-standard-deviation principle in statistical anomaly detection. In clinical behavior analysis and data statistics, fluctuations in the daily functional status of the elderly usually approximate a normal distribution. If the variance deviation exceeds 2.0 (i.e., the current feature value deviates from the individual's historical mean by more than two standard deviations), statistically, this is a low-probability event with a probability of less than 5%. This means that the data deviation at this time is highly likely not a reasonable random fluctuation in the subject's daily status, but rather reflects a real functional decline with clinical statistical significance or a psychological deterioration due to extreme fear of falling; conversely, if the variance deviation is less than this threshold, it indicates that the current dependent behavior is still within the individual's normal physiological fluctuation range. In actual system deployment, the threshold can also be dynamically adjusted between 1.5 (high-sensitivity early screening) and 3.0 (high-specificity diagnosis) according to the sensitivity requirements of fall prevention warnings in different care scenarios.

[0164] Exemplary system:

[0165] Please see Figure 2 A home-based elderly care image assessment system for daily activity ability includes:

[0166] The data acquisition module is used to acquire continuous temporal images of home scenes and perform feature analysis to construct a physical boundary model containing surface normal vectors, and extract the three-dimensional coordinates of the centroid and the three-dimensional coordinates of the upper limb end of the human body pose in the continuous temporal sequence.

[0167] The kinematics calculation module is used to calculate the centroid velocity vector based on the three-dimensional coordinates of the centroid and the upper limb velocity vector based on the three-dimensional coordinates of the upper limb distal end, and simultaneously calculate the shortest spatial distance from the three-dimensional coordinates of the upper limb distal end to the physical boundary model;

[0168] The temporal sequence division module is used to divide the continuous temporal sequence into effective supported frames and unsupported free frames based on the shortest spatial distance, the centroid velocity vector, the upper limb velocity vector, and the surface normal vector.

[0169] The initial feature generation module is used to calculate the trajectory bias cost between the actual trajectory distance and the ideal straight line distance based on the three-dimensional coordinates of the centroid, and to perform a fusion calculation on the number of effective support frames and the trajectory bias cost using a preset initial weight coefficient to obtain an initial environmental support feature value that characterizes the explicit physical compensation tendency.

[0170] The feedback adjustment module is used to determine whether the initial environmental support feature value is greater than a preset trigger threshold; if so, it extracts the centroid three-dimensional coordinates corresponding to the unsupported free frame, calculates the centroid fluctuation variance of the unsupported free frame in a preset vertical direction and maps it to gait stability confidence; it adjusts the initial weight coefficients based on the gait stability confidence, and re-executes the fusion calculation through the adjusted weight coefficients to generate the corrected environmental support feature value;

[0171] An anomaly warning module is used to compare the corrected environmental feature values ​​with a preset historical baseline to obtain the variation deviation. If the variation deviation is greater than a preset variation threshold, a degradation warning signal is generated.

[0172] Exemplary computer-readable medium:

[0173] Embodiments of this application may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps described in the "Exemplary Methods" section above according to the various embodiments of this application.

[0174] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0175] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0176] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0177] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0178] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0179] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for image-based assessment of the daily activity abilities of elderly people living at home, characterized in that, Includes the following steps: Acquire continuous temporal images of home scenes and perform feature analysis to construct a physical boundary model containing surface normal vectors, and extract the three-dimensional coordinates of the centroid and the three-dimensional coordinates of the upper limb end in the continuous temporal sequence. The centroid velocity vector is calculated based on the three-dimensional coordinates of the centroid, and the upper limb velocity vector is calculated based on the three-dimensional coordinates of the upper limb distal end. At the same time, the shortest spatial distance from the three-dimensional coordinates of the upper limb distal end to the physical boundary model is calculated. Based on the shortest spatial distance, the centroid velocity vector, the upper limb velocity vector, and the surface normal vector, the continuous temporal sequence is divided into effective supported frames and unsupported free frames; The trajectory bias cost between the actual trajectory distance and the ideal straight line distance is calculated based on the three-dimensional coordinates of the centroid. The number of effective support frames and the trajectory bias cost are fused and calculated using a preset initial weight coefficient to obtain an initial environmental support feature value that characterizes the explicit physical compensation tendency. Determine whether the initial environment-dependent feature value is greater than a preset trigger threshold; If so, extract the three-dimensional coordinates of the centroid corresponding to the unsupported free frame, calculate the centroid fluctuation variance of the unsupported free frame in the preset vertical direction and map it into gait stability confidence. The initial weight coefficients are adjusted based on the gait stability confidence, and the fusion calculation is re-executed using the adjusted weight coefficients to generate corrected environmental dependency feature values. The corrected environmental feature value is compared with a preset historical baseline to obtain the variation deviation. If the variation deviation is greater than a preset variation threshold, a decline warning signal is generated.

2. The image-based assessment method for the daily activity abilities of elderly people living at home according to claim 1, characterized in that, The process of dividing the continuous temporal sequence into supported frames and unsupported free frames based on the shortest spatial distance, the centroid velocity vector, the upper limb velocity vector, and the surface normal vector includes the following steps: Determine whether the shortest spatial distance is less than or equal to a preset distance threshold; If not, the current frame is classified as an unattached, free-floating frame; If so, and it satisfies any of the following conditions: If the magnitude of the centroid velocity vector is less than or equal to a preset static velocity threshold, or if the centroid velocity vector, the upper limb velocity vector, and the surface normal vector satisfy a preset dynamic cooperative sliding condition, then the current frame is classified as a valid support frame.

3. The image-based assessment method for the daily activity abilities of elderly people living at home according to claim 2, characterized in that, The centroid velocity vector, the upper limb velocity vector, and the surface normal vector satisfy the preset dynamic cooperative sliding condition, including the following steps: Based on the upper limb velocity vector and the surface normal vector, calculate the tangential velocity component of the upper limb velocity vector on the surface of the physical boundary model; Calculate the ratio of the magnitude of the tangential velocity component to the magnitude of the upper limb velocity vector, and calculate the spatial angle between the upper limb velocity vector and the center of mass velocity vector; When the ratio of tangential velocities is greater than a preset tangential ratio threshold and the spatial angle is less than a preset angle threshold, it is determined that the upper limb velocity vector and the surface normal vector satisfy the dynamic cooperative sliding condition.

4. The image-based assessment method for the daily activity abilities of elderly people living at home according to claim 1, characterized in that, The calculation of the trajectory bias cost between the actual trajectory distance and the ideal straight-line distance based on the three-dimensional coordinates of the centroid includes the following steps: Obtain the first centroid three-dimensional coordinates corresponding to the start frame and the second centroid three-dimensional coordinates corresponding to the end frame of the continuous time sequence from the centroid three-dimensional coordinates, and calculate the ideal straight-line distance between the first centroid three-dimensional coordinates and the second centroid three-dimensional coordinates; The centroid three-dimensional coordinates of adjacent frames in the continuous time sequence are traversed, and the coordinate displacement between adjacent frames is accumulated to obtain the actual trajectory distance; The difference between the actual trajectory distance and the ideal straight line distance is calculated, and the ratio of the difference to the ideal straight line distance is determined as the trajectory bias cost.

5. The image-based assessment method for the daily activity abilities of elderly people living at home according to claim 4, characterized in that, The process of traversing the three-dimensional coordinates of the centroid of adjacent frames in the continuous time sequence and accumulating the coordinate displacement between adjacent frames to obtain the actual trajectory distance includes the following steps: Extract the two-dimensional projection coordinates of the centroid three-dimensional coordinates on the horizontal plane corresponding to each of the adjacent frames; The two-dimensional projected coordinates are denoised using a preset smoothing filtering algorithm to obtain smoothed two-dimensional projected coordinates. Calculate and sum the Euclidean distances between adjacent smoothed 2D projected coordinates, and determine the summed result as the actual trajectory distance.

6. The image-based assessment method for the daily activity abilities of elderly people living at home according to claim 1, characterized in that, Obtaining the initial environmental dependency characteristic value that characterizes the dominant physical compensation tendency includes the following steps: Obtain the total number of frames in the continuous time sequence, calculate the ratio of the number of effective dependent frames to the total number of frames, and obtain the normalized dependent frame rate; Extract the preset first initial weight coefficient and second initial weight coefficient; The first feature term is obtained by multiplying the dependent frame rate by the first initial weight coefficient, and the second feature term is obtained by multiplying the trajectory bias cost by the second initial weight coefficient. The first feature term and the second feature term are summed to obtain the initial environmental dependency feature value.

7. The image-based assessment method for the daily activity abilities of elderly people living at home according to claim 1, characterized in that, The calculation of the centroid fluctuation variance of the unsupported free frame in the preset vertical direction and its mapping to gait stability confidence includes the following steps: Extract the vertical coordinate components of the centroid three-dimensional coordinates in the preset vertical direction corresponding to each of the aforementioned unsupported free frames; Calculate the mean of all the vertical coordinate components, and calculate the variance of all the vertical coordinate components based on the mean to obtain the centroid fluctuation variance; Calculate the ratio of the preset standard fluctuation variance to the centroid fluctuation variance, and generate a gait stability confidence score based on the ratio using a preset mapping function; wherein the gait stability confidence score is negatively correlated with the centroid fluctuation variance.

8. The image-based assessment method for the daily activity abilities of elderly people living at home according to claim 6, characterized in that, The process of generating the corrected environment-dependent feature value includes the following steps: Extract all the valid support frames in the continuous time series, and identify all the continuous support frame segments in chronological order; Obtain the frame length of each consecutive supporting frame segment, filter out the segment with the longest frame length, and calculate the ratio of the longest frame length of the segment to the number of effective supporting frames to obtain the supporting time continuity index. The gait stability confidence and the time continuity index are input into a preset nonlinear mapping function; When the gait stability confidence is less than or equal to the preset confidence threshold, a positive gain adjustment factor greater than 1 is output. When the gait stability confidence is greater than the preset confidence threshold and the reliance time continuity index is less than the preset continuity threshold, a negative decay adjustment factor less than 1 is output. When the gait stability confidence is greater than the preset confidence threshold and the reliance time continuity index is greater than or equal to the preset continuity threshold, a non-monotonic inversion mechanism is triggered, and a higher-order gain adjustment factor greater than 1 is output in the reverse direction. The adjusted first weight coefficient is obtained by multiplying the output adjustment factor by the first initial weight coefficient. The adjusted first weight coefficient is used to perform reverse compensation on the second initial weight coefficient to obtain the adjusted second weight coefficient. Using the adjusted first weight coefficient and the adjusted second weight coefficient, the dependent frame rate and the trajectory bias cost are summed again to generate the corrected environment dependent feature value.

9. A system for implementing the image-based assessment method for the daily activity abilities of elderly people living at home according to any one of claims 1-8, characterized in that, include: The data acquisition module is used to acquire continuous temporal images of home scenes and perform feature analysis to construct a physical boundary model containing surface normal vectors, and extract the three-dimensional coordinates of the centroid and the three-dimensional coordinates of the upper limb end of the human body pose in the continuous temporal sequence. The kinematics calculation module is used to calculate the centroid velocity vector based on the three-dimensional coordinates of the centroid and the upper limb velocity vector based on the three-dimensional coordinates of the upper limb distal end, and simultaneously calculate the shortest spatial distance from the three-dimensional coordinates of the upper limb distal end to the physical boundary model; The temporal sequence division module is used to divide the continuous temporal sequence into effective supported frames and unsupported free frames based on the shortest spatial distance, the centroid velocity vector, the upper limb velocity vector, and the surface normal vector. The initial feature generation module is used to calculate the trajectory bias cost between the actual trajectory distance and the ideal straight line distance based on the three-dimensional coordinates of the centroid, and to perform a fusion calculation on the number of effective support frames and the trajectory bias cost using a preset initial weight coefficient to obtain an initial environmental support feature value that characterizes the explicit physical compensation tendency. The feedback adjustment module is used to determine whether the initial environment support feature value is greater than the preset trigger threshold; if so, the three-dimensional coordinates of the centroid corresponding to the unsupported free frame are extracted, the centroid fluctuation variance of the unsupported free frame in the preset vertical direction is calculated and mapped to gait stability confidence. The initial weight coefficients are adjusted based on the gait stability confidence, and the fusion calculation is re-executed using the adjusted weight coefficients to generate the corrected environmental dependency feature values. An anomaly warning module is used to compare the corrected environmental feature values ​​with a preset historical baseline to obtain the variation deviation. If the variation deviation is greater than a preset variation threshold, a degradation warning signal is generated.

10. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method as described in any one of claims 1-8.

Images