Walking assisting nursing device for neurosurgery nursing

Abstract

The invention relates to the technical field of auxiliary walking devices, and particularly discloses an auxiliary walking nursing device for neurosurgical nursing, which is characterized in that an inertial measurement unit (IMU) and a pressure sensor are used for respectively acquiring three-dimensional motion data and standing pressure data of a patient object; a neural network model based on deep learning is introduced to carry out data analysis on the three-dimensional motion data and the standing pressure data of the patient object, so that the time sequence correlation change characteristics of the motion state of the patient object and the dynamic change trend of the standing pressure are mined, and then fusion analysis is carried out on the time sequence correlation change characteristics and the dynamic change trend; therefore, intelligent identification of abnormal actions of a patient object is realized, and an early warning prompt is automatically generated and a built-in safety mechanism is started to lock a position when the abnormal actions occur. In this way, active safety protection of the patient object can be achieved, the safety and intelligent level of the auxiliary walking nursing device are effectively improved, and then the safety and efficiency of rehabilitation training are improved.

Description

Technical Field

[0001] The present application relates to the technical field of auxiliary walking devices, and more specifically, to an auxiliary walking nursing device for neurosurgery nursing. Background Art

[0002] In the field of neurosurgery nursing, assisted walking devices are crucial to help postoperative patients recover their walking ability and prevent complications caused by long-term bed rest. At present, most of the existing assisted walking nursing devices for neurosurgery nursing are debugged by medical staff and then tried by patients. They cannot be adjusted most appropriately according to the patient's own usage experience, which brings inconvenience to users. In this regard, the invention patent with publication number CN114177013A proposes an assisted walking nursing device for neurosurgery nursing, which includes two groups of support mechanisms, each of which includes a support column, a bracket and two legs, and rollers are installed at the bottom of the two legs. Adjustment mechanisms are installed on the two support columns, allowing users to adjust the height of the armrests by themselves through simple mechanical structures (such as sliding sleeves, operating chambers, latches, springs, etc.), thereby meeting the needs of patients with different heights and usage scenarios.

[0003] In addition, in order to ensure the safety of patients, fall prevention and emergency response are of great significance during the rehabilitation training of patients. However, in the prior art, although the assisted walking care device is equipped with an alarm and a control panel, this passive response mechanism requires the user to manually trigger the alarm or help button when encountering a problem, and cannot actively sense the user's movement state for early warning, which may lead to delayed response and increase the risk of accidents.

[0004] Therefore, an optimized walking assistance nursing device for neurosurgery nursing is expected. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an auxiliary walking care device for neurosurgery care, which uses an inertial measurement unit (IMU) and a pressure sensor to respectively collect the three-dimensional motion data and standing pressure data of the patient object, and introduces a neural network model based on deep learning to perform data analysis on the three-dimensional motion data and standing pressure data of the patient object, so as to dig out the time-series correlation change characteristics of the patient object's motion state and the dynamic change trend of the standing pressure, and then through the fusion analysis of the two, to achieve intelligent recognition of abnormal movements of the patient object, and automatically generate early warning prompts when abnormal movements occur and start the built-in safety mechanism to lock the position. In this way, active safety protection of the patient object can be achieved, effectively improving the safety and intelligence level of the auxiliary walking care device, and thus helping to improve the safety and efficiency of rehabilitation training.

[0006] Accordingly, in one aspect of the present application, there is provided an assisted walking care device for neurosurgical care, which includes:

[0007] Two groups of support mechanisms, each group of support mechanisms respectively includes a support column, a bracket and two legs. The bracket is fixedly connected to the lower end of the support column, the two legs are fixedly connected to the lower end of the bracket, and rollers are installed at the bottoms of the two legs;

[0008] The brackets in the two groups of support mechanisms are fixedly connected by a connecting frame, and cross bars are fixedly installed at the upper ends of the two support columns;

[0009] Controllers are respectively installed on one side of the two support columns away from each other, which are used to adjust the height of the armrest and fix the position of the armrest through a mechanical locking mechanism.

[0010] In the above-mentioned assisted walking care device for neurosurgical care, the controller includes: a motion data acquisition module, which is used to obtain the time queue of the three-dimensional motion data of the patient object collected by the IMU, and the time queue of the standing pressure value collected by the pressure sensor; a motion state feature extraction module, which is used to extract the motion state features of the patient at each time point based on the time queue of the three-dimensional motion data to obtain the time queue of the single-point motion state fully connected embedding coding vectors; a motion state time-series context awareness module, which is used to perform time-series context association coding based on the motion state change perception on the time queue of the single-point motion state fully connected embedding coding vectors to obtain the motion state time-series context association coding vectors; a standing pressure time-series pattern feature extraction module, which is used to extract the standing pressure time-series pattern features of the time queue of the standing pressure value to obtain the standing pressure time-series context association coding vectors; a patient action anomaly recognition module, which is used to determine the detection result of the patient's action state based on the fusion features of the standing pressure time-series context association coding vectors and the motion state time-series context association coding vectors; a safety locking module, which is used to generate a warning prompt and start the built-in safety mechanism to lock the position in response to the detection result that there is an anomaly in the patient object's action.

[0011] Compared with the prior art, the auxiliary walking nursing device for neurosurgical care provided by the present application uses an inertial measurement unit (IMU) and a pressure sensor to collect three-dimensional motion data and standing pressure data of a patient object respectively, and introduces a neural network model based on deep learning to perform data analysis on the three-dimensional motion data and standing pressure data of the patient object, so as to extract the time-series correlation change characteristics of the patient object's motion state and the dynamic change trend of the standing pressure. Furthermore, through the fusion analysis of the two, the intelligent recognition of the abnormal actions of the patient object can be realized, and an early warning prompt can be automatically generated and the built-in safety mechanism can be activated to lock the position when an abnormal action occurs. In this way, the active safety protection of the patient object can be realized, the safety and intelligent level of the auxiliary walking nursing device can be effectively improved, and thus the safety and efficiency of the rehabilitation training can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other objects, features and advantages of the present application will become more obvious. The accompanying drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application, and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0013] Figure 1 It is a block diagram of an auxiliary walking nursing device for neurosurgical care according to an embodiment of the present application.

[0014] Figure 2 It is a schematic diagram of data flow of an auxiliary walking nursing device for neurosurgical care according to an embodiment of the present application.

[0015] Figure 3 It is a block diagram of a motion state time-series context awareness module of an auxiliary walking nursing device for neurosurgical care according to an embodiment of the present application.

[0016] Figure 4 It is a block diagram of a patient action abnormal recognition module of an auxiliary walking nursing device for neurosurgical care according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] As shown in the present application and the claims, unless the context clearly indicates an exception, words such as "a", "an", "one" and / or "the" are not specifically singular, but may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0018] Although this application makes various references to certain modules in the system according to embodiments of this application, any number of different modules can be used and run on a user terminal and / or a server. These modules are merely illustrative, and different aspects of the system and method can use different modules.

[0019] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the operations before or below do not necessarily have to be executed precisely in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more steps can be removed from these processes.

[0020] Next, example embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all embodiments of this application. It should be understood that this application is not limited by the example embodiments described herein.

[0021] It is worth noting that in this application, all actions of obtaining data are carried out on the premise of complying with the corresponding data protection regulations and policies of the country where it is located and obtaining authorization from the owner of the corresponding device.

[0022] As mentioned in the above background art, patent CN114177013A proposes an auxiliary walking care device for neurosurgical care, which includes two sets of support mechanisms. Each set of support mechanisms respectively includes a support column, a bracket, and two legs. The bracket is fixedly connected to the lower end of the support column, the two legs are fixedly connected to the lower end of the bracket, and rollers are installed at the bottom of both legs; the brackets in the two sets of support mechanisms are fixedly connected by a connecting frame, and crossbars are fixedly installed at the upper ends of both support columns; controllers are respectively installed on one side of the two support columns away from each other, for adjusting the height of the armrest and fixing the position of the armrest through a mechanical locking mechanism.

[0023] To ensure the safety of patients, during the rehabilitation training, preventing falls and a rapid response to emergencies are of crucial importance. However, in the existing technologies, although the assisted walking device is equipped with an alarm system and a control interface, these facilities rely on the user to manually activate the alarm or the SOS button when encountering difficulties, which belongs to a post-reaction mode and cannot automatically monitor the user's movement status and give a warning in advance. This may lead to a lag in response and increase the risk of accidents. To address the above technical problems, the present application proposes an optimized assisted walking care device for neurosurgical nursing, which uses an inertial measurement unit (IMU) and a pressure sensor to collect the three-dimensional motion data and standing pressure data of the patient object respectively, and introduces a neural network model based on deep learning to analyze the three-dimensional motion data and standing pressure data of the patient object, so as to extract the time-series correlation change characteristics of the patient object's motion state and the dynamic change trend of the standing pressure. Furthermore, through the fusion analysis of the two, the intelligent recognition of the abnormal actions of the patient object is realized, and a warning prompt is automatically generated and the built-in safety mechanism is activated to lock the position when the abnormal action occurs. In this way, the active safety protection of the patient object can be achieved, effectively improving the safety and intelligence level of the assisted walking care device, and thus contributing to improving the safety and efficiency of the rehabilitation training.

[0024] Figure 1 FIG. is a block diagram of an assisted walking care device for neurosurgical nursing according to an embodiment of the present application. Figure 2 FIG. is a schematic diagram of the data flow of an assisted walking care device for neurosurgical nursing according to an embodiment of the present application. As Figure 1 and Figure 2As shown, the auxiliary walking care device 100 for neurosurgical care includes: a motion data acquisition module 110, configured to obtain a time series of three-dimensional motion data of a patient object collected by an IMU, and a time series of standing pressure values collected by a pressure sensor; a motion state feature extraction module 120, configured to extract motion state features of the patient at each time point based on the time series of the three-dimensional motion data to obtain a time series of single-point motion state fully connected embedding coding vectors; a motion state time-series context awareness module 130, configured to perform time-series context association coding based on motion state change awareness on the time series of the single-point motion state fully connected embedding coding vectors to obtain motion state time-series context association coding vectors; a standing pressure time-series pattern feature extraction module 140, configured to extract standing pressure time-series pattern features of the time series of the standing pressure values to obtain standing pressure time-series context association coding vectors; a patient action anomaly recognition module 150, configured to determine a detection result of the patient's action state based on the fusion features of the standing pressure time-series context association coding vectors and the motion state time-series context association coding vectors; and a safety locking module 160, configured to generate a warning prompt and activate a built-in safety mechanism to lock the position in response to the detection result indicating that there is an anomaly in the patient object's action.

[0025] In the above-mentioned auxiliary walking care device for neurosurgical care, the motion data acquisition module 110 is configured to obtain a time series of three-dimensional motion data of a patient object collected by an IMU, and a time series of standing pressure values collected by a pressure sensor. In a specific example of the present application, the three-dimensional motion data includes X-axis acceleration, Y-axis acceleration, Z-axis acceleration, X-axis angular velocity, Y-axis angular velocity, and Z-axis angular velocity. Specifically, an inertial measurement unit (IMU) integrates an accelerometer and a gyroscope inside, and can measure the acceleration and angular velocity of a patient object in three axes (X-axis, Y-axis, Z-axis) in real time, so as to comprehensively reflect the motion state of the patient object. Among them, the accelerometer measures the acceleration of an object in the X, Y, and Z axes by detecting the force generated by the mass block under the action of acceleration according to Newton's second law; the gyroscope measures the angular velocity of an object around three axes by detecting the Coriolis force using the principle of conservation of angular momentum. The pressure sensor is used to monitor the pressure value of the patient on the ground during standing in real time. Based on principles such as piezoresistive effect and piezoelectric effect, it converts the pressure applied by the patient on the sensor during standing into an electrical signal, and then outputs the corresponding pressure value, which can reflect information such as the standing stability of the patient. By comprehensively analyzing the three-dimensional motion data and standing pressure data of the patient object, the present application can comprehensively understand the body dynamics and force conditions of the patient during walking, thereby providing indispensable data support for subsequent early warning of abnormal behaviors.

[0026] Specifically, an IMU is a device capable of detecting changes in acceleration and angular velocity in three-dimensional space, usually including a three-axis accelerometer and a three-axis gyroscope. For the selection of an IMU, considering the application scenario of the assisted walking care device and the characteristics of the target user group, sensors with high sensitivity and low noise characteristics need to be selected. Such sensors can accurately respond to slight movement changes of the patient and can filter out unnecessary environmental interference signals. For example, for the irregular movement patterns that may exist in patients after neurosurgery, the selected IMU needs to be able to capture subtle acceleration and angular velocity changes while maintaining long-term stable operation.

[0027] Specifically, the IMU should be installed at key positions that can represent the overall body movement, such as the patient's waist or ankle. The waist, being the center of the body's gravity, can reflect most of the trunk movement characteristics; while the ankle is directly related to walking and can provide important information about gait. In some cases, additional IMUs can also be considered to be set on the armrest of the assisted walking device to monitor the arm movements, so as to obtain more comprehensive posture information. No matter which installation position is chosen, it is necessary to ensure that the IMU is firmly fixed to avoid data distortion caused by loosening.

[0028] Since rollers are installed at the bottom of the legs of the support structure, pressure sensors are placed here to directly measure the pressure of the rollers on the ground. To improve the measurement accuracy, multiple pressure sensors can be configured at the bottom of each leg to form a small array. This layout not only increases the number of data points but also can judge the center of gravity distribution of the patient when standing by analyzing the differences between the sensors. In addition, considering the large weight differences among different patients, the pressure sensors need to have a wide measurement range, which can not only meet the usage requirements of lightweight users but also withstand the pressure impact brought by heavyweight users.

[0029] The data recorded by the IMU includes the acceleration in the X-axis, Y-axis, and Z-axis directions and the angular velocity in the corresponding axial directions. These data reflect the changes in the body's center of gravity when the patient is standing and walking, as well as the speed changes in the body's rotation angle. The acceleration and angular velocity values at each moment together form a six-dimensional vector, representing the patient's motion state at that moment. As time goes by, the IMU continuously generates such a sequence of six-dimensional vectors, forming a time queue of three-dimensional motion data that evolves over time.

[0030] Meanwhile, to supplement the IMU data and further understand the patient's postural stability during standing, pressure sensors are integrated into the leg parts of the support structure. When the patient stands or walks, the pressure exerted by the feet on the ground is transmitted to the rollers and finally captured by the pressure sensors at the bottom of the legs. These pressure sensors can sense and quantify the force distribution generated when both feet contact the ground. As the patient moves and the standing position changes, the pressure sensors continuously record a series of standing pressure values, which are also arranged in chronological order to form a time queue of standing pressure values.

[0031] For both the IMU and pressure sensors, ensuring high-frequency sampling is crucial to capture as many movement details as possible, providing richer raw materials for subsequent analysis. At the same time, considering data transmission efficiency and processing speed, the sampling rate needs to be set reasonably, neither too high to cause an excessive burden on the system nor too low to miss key movement features. Generally, according to clinical application requirements, a suitable sampling frequency is selected to meet both real-time requirements and ensure data quality.

[0032] In addition, to ensure data accuracy, both the IMU and pressure sensors need to go through a strict calibration procedure. Due to factors such as manufacturing process differences, there may be slight deviations in products of different batches. Therefore, during the calibration stage, a standard test environment is used to adjust the sensor output to ensure that all devices work on the same baseline. Moreover, considering possible interference factors in the actual use environment, such as magnetic field effects, corresponding compensation measures are also taken to reduce the impact of external conditions on the data acquisition accuracy.

[0033] During the data collection process, to facilitate subsequent processing, the raw data generated by the IMU and pressure sensors are timestamped to mark the specific collection moment. This not only helps to establish a time series relationship but also facilitates comparative analysis with other simultaneously collected data sources. For example, if the assisted walking device is equipped with video monitoring or other types of sensors, precise alignment of multi-source data can be achieved through timestamps, and then a more complete patient behavior model can be constructed.

[0034] In the above-mentioned auxiliary walking nursing device for neurosurgical care, the motion state feature extraction module 120 is used to extract the motion state features of the patient at each time point based on the time queue of the three-dimensional motion data, so as to obtain the time queue of the single-point motion state fully-connected embedding coding vectors. In a specific example of the present application, the motion state feature extraction module 120 is used to: perform fully-connected embedding coding on each three-dimensional motion data in the time queue of the three-dimensional motion data by using a single-point motion state embedding encoder based on a fully-connected layer to obtain the time queue of the single-point motion state fully-connected embedding coding vectors. It should be understood that since the information of each dimension such as the X-axis acceleration, Y-axis acceleration, Z-axis acceleration, X-axis angular velocity, Y-axis angular velocity, and Z-axis angular velocity in the three-dimensional motion data is interrelated and jointly constitutes a complete description of the motion state of the patient object. Therefore, in order to effectively extract the motion state features of the patient object at each time point, the present application uses a single-point motion state embedding encoder based on a fully-connected layer to perform fully-connected embedding coding on the three-dimensional motion data at each time point respectively. Those of ordinary skill in the art should know that each neuron in the fully-connected layer is connected to all dimensions of the input data, and by performing non-linear transformation and combination on the information of each dimension in the three-dimensional motion data, complex patterns in the data are learned, and the internal connections between different dimension information are captured, so that the time queue of the original three-dimensional motion data can be converted into a time queue of high-level and more expressive single-point motion state fully-connected embedding coding vectors, so as to more accurately describe the motion state of the patient object at each time step.

[0035] In the above-mentioned auxiliary walking nursing device for neurosurgical care, the motion state temporal context awareness module 130 is used to perform temporal context correlation coding based on motion state change awareness on the time queue of the single-point motion state fully-connected embedding coding vectors to obtain motion state temporal context correlation coding vectors. In particular, considering that the patient's motion is a continuous process and there is a correlation between the motion states at the previous and subsequent moments. Some abnormal actions may not be formed instantaneously, but are the result of gradual evolution over a period of time. Therefore, in order to effectively capture the change trend and context connection of the patient's motion state in the time dimension, the present application proposes a temporal context correlation coding method based on neighborhood motion state change awareness, which adaptively adjusts the way and intensity of feature transfer between the patient's motion state information at each time step by analyzing the change of the motion state between adjacent time steps, so as to more accurately understand the temporal change trend of the patient's motion state. Among them, Figure 3 is a block diagram of the motion state temporal context awareness module of the auxiliary walking nursing device for neurosurgical care according to an embodiment of the present application. As Figure 3As shown, the motion state time-series context awareness module 130 includes: a feature transfer significance measurement unit 131, configured to perform feature transfer significance measurement based on neighborhood motion state change awareness on each single-point motion state fully connected embedding coding vector in the time queue of the single-point motion state fully connected embedding coding vectors to obtain a time queue of single-point motion state feature transfer significant factors; and a context aggregation coding unit 132, configured to perform feature attention modulation context aggregation coding on the time queue of the single-point motion state fully connected embedding coding vectors based on the time queue of the single-point motion state feature transfer significant factors to obtain the motion state time-series context association coding vector.

[0036] Specifically, in a specific example of the present application, the feature transfer significance measurement unit 131 is configured to: calculate the feature jump degree of each single-point motion state fully connected embedding coding vector in the time queue of the single-point motion state fully connected embedding coding vectors to obtain a time queue of single-point motion state feature jump degrees, which is expressed by the formula:

[0037] J = {v 1 , v 2 ,..., v i ,..., v n}

[0038]

[0039] where J represents the time queue of the single-point motion state fully connected embedding coding vectors, v 1 , v 2 , v i and v n respectively represent the 1st, 2nd, i-th, and n-th single-point motion state fully connected embedding coding vectors in the time queue of the single-point motion state fully connected embedding coding vectors, n is the number of feature vectors in the time queue of the single-point motion state fully connected embedding coding vectors, represents the feature value at the j-th position in the i-th single-point motion state fully connected embedding coding vector, L is the feature scale value of the i-th single-point motion state fully connected embedding coding vector, u i is the feature energy intensity factor of the i-th single-point motion state fully connected embedding coding vector, u i+1 represents the feature energy intensity factor of the (i + 1)-th single-point motion state fully connected embedding coding vector, and t i represents the feature jump degree of the i-th single-point motion state fully connected embedding coding vector.

[0040] That is, considering the dynamic characteristics of the patient's movement and the possible differences in the speed and amplitude of the movement state changes at different stages, in order to accurately capture the actual changes in the patient's movement state at each time node, the present application adopts a sliding window method to traverse the fully connected embedding coding vectors of the single-point movement states at all adjacent time nodes. By calculating the feature jump degree between these adjacent coding vectors, the degree of change in the movement state features of each time node relative to the neighboring time points can be quantified. It should be understood that if the feature jump degree of a single-point movement state fully connected embedding coding vector relative to its adjacent time node is large, it indicates that the movement state of the patient at this time node has changed significantly, and there may be situations such as a fall risk. At this time, the movement state features of this time node need to be focused on and analyzed.

[0041] Specifically, in a specific example of the present application, the feature transfer significance measurement unit 131 is further configured to: calculate the feature transfer space span of each single-point movement state fully connected embedding coding vector in the time queue of the single-point movement state fully connected embedding coding vector to obtain a time queue of the single-point movement state feature transfer space span, which is represented by the formula:

[0042] s i = Count(v i → v n )

[0043] where s i represents the feature transfer space span of the i-th single-point movement state fully connected embedding coding vector, and Count(v i → v n ) represents the number of feature vectors separated between v i and v n .

[0044] That is, considering that the influence of the motion state characteristics at the early time node on the motion state at the current time point may weaken as the time span increases, that is, there is an information transfer time series attenuation effect. Therefore, when performing time series context encoding on the time queue of the single-point motion state fully connected embedding encoding vector, the present application further calculates the feature transfer spatial span of the single-point motion state fully connected embedding encoding vector at each time step relative to the current time step. In order to accurately evaluate the importance of features at different time steps, for those motion state features with a longer time span and a larger feature transfer spatial span relative to the current time step, the system assigns a smaller weight; while for those motion state features with a shorter time span and a smaller feature transfer spatial span, a larger weight is assigned. In this way, the weights of features at different time points are dynamically adjusted, so as to better capture the influence of the recent motion state on the current state, enhance the sensitivity of the model to recent changes, and more reasonably reflect the information importance of each single-point motion state fully connected embedding encoding vector.

[0045] Specifically, in a specific example of the present application, the feature transfer significance measurement unit 131 is further configured to: calculate the feature transfer significant factor of each single-point motion state fully connected embedding encoding vector based on the feature jump degree and the feature transfer spatial span of each single-point motion state fully connected embedding encoding vector in the time queue of the single-point motion state fully connected embedding encoding vector, so as to obtain the time queue of the single-point motion state feature transfer significant factor, which is expressed by the formula:

[0046]

[0047] where α and β are preset weight parameters used to balance the influence of the feature jump degree and the feature transfer spatial span, and e i represents the feature transfer significant factor of the i-th single-point motion state fully connected embedding encoding vector.

[0048] That is, in order to more comprehensively evaluate the importance of each single-point motion state fully connected embedding encoding vector in the context information transfer, the present application comprehensively considers two key factors: the feature jump degree and the feature transfer spatial span. By combining the information in these two aspects, the feature transfer significant factor is calculated, so as to highlight the feature information of important time nodes and reduce the interference of noise information in the subsequent motion state feature context transfer encoding process, thereby improving the accuracy and efficiency of perceiving the time series change of the patient's overall motion state.

[0049] Specifically, in a specific example of the present application, the context aggregation encoding unit 132 is configured to: input the time queue of the single-point motion state feature transfer significant factor into a gated transfer unit including a softmax normalization function to obtain a time queue of the single-point motion state feature transfer significant weight, which is expressed by the formula:

[0050]

[0051] where softmax(·) is the normalization exponential function, mask[·] is the gated mask function, τ is the gated threshold, and w i is the single-point motion state feature transfer significant weight of the v i ;

[0052] Based on the time queue of the single-point motion state feature transfer significant weight, perform weighted aggregation on the time queue of the single-point motion state fully connected embedding encoding vector to obtain the motion state temporal context correlation encoding vector, which is expressed by the formula:

[0053]

[0054] where v f represents the motion state temporal context correlation encoding vector.

[0055] That is, use the gated mechanism to perform gated screening on the feature transfer significant factors of each single-point motion state fully connected embedding encoding vector, generate a time queue of the single-point motion state feature transfer significant weight, and use this to perform weighted aggregation on the original time queue of the single-point motion state fully connected embedding encoding vector, so as to obtain the motion state temporal context correlation encoding vector. In this way, not only the patient motion state information of each time node is effectively fused, but also the temporal decay effect of information transfer and the significant change of the patient motion state are taken into account, thus realizing a more accurate and comprehensive feature description of the patient's overall motion state.

[0056] In the above-mentioned auxiliary walking nursing device for neurosurgery nursing, the standing pressure time series pattern feature extraction module 140 is used to extract the standing pressure time series pattern features of the time queue of the standing pressure value to obtain the standing pressure time series context associated coding vector. In a specific example of the present application, the standing pressure time series pattern feature extraction module 140 is used to: input the time queue of the standing pressure value into the standing pressure time series pattern feature extractor based on the forward LSTM model to obtain the standing pressure time series context associated coding vector. It should be understood that the patient's standing pressure will change dynamically with factors such as movement state and body posture. Therefore, in order to effectively capture the time series change pattern of the patient's standing pressure to reveal the patient's movement stability, the present application uses a forward LSTM model to perform time series encoding on the time queue of the standing pressure value, so as to use the long-term memory capacity of the LSTM model to mine the patient's standing pressure time series fluctuation characteristics and generate a standing pressure time series context associated coding vector. Those skilled in the art should know that the forward LSTM model is a neural network architecture suitable for processing and predicting long-term dependencies in time series data. It can capture long-term dependencies and short-term dynamic changes through a special memory unit structure, and is very suitable for analyzing time-related data series such as standing pressure values. When the time series of standing pressure values ​​is input, the forget gate in the model determines which information should be retained or forgotten, the input gate is combined with the candidate cell state to update the cell state, and the output gate adjusts the final output. This mechanism enables the forward LSTM to learn the pattern of standing pressure evolution over time, identify potential rules, and generate a standing pressure time series context-related encoding vector.

[0057] In the above-mentioned walking assistance nursing device for neurosurgery nursing, the patient movement abnormality recognition module 150 is used to determine the detection result of the patient's movement state based on the fusion features of the standing pressure temporal context association coding vector and the motion state temporal context association coding vector. Figure 4 FIG. 1 is a block diagram of a patient movement abnormality recognition module of an auxiliary walking care device for neurosurgery care according to an embodiment of the present application. Figure 4 As shown, the patient motion abnormality recognition module 150 includes: a cascade fusion unit 151, used for cascading the standing pressure temporal context association coding vector and the motion state temporal context association coding vector to obtain a motion state multimodal cascade coding vector; a classification recognition unit 152, used for inputting the motion state multimodal cascade coding vector into a classifier-based abnormal behavior detector to obtain the detection result, and the detection result is used to indicate whether the patient object's motion is abnormal.

[0058] Specifically, the cascade fusion unit 151 is configured to cascade the standing pressure time-series context-associated encoding vector and the motion state time-series context-associated encoding vector to obtain a motion state multi-modal cascade encoding vector. It should be understood that the standing pressure data can reflect the pressure distribution and changes when the patient's feet contact the ground, and can reflect information such as the transfer of the body center of gravity and the balance state; while the motion state data includes information such as acceleration and angular velocity, and can reflect the body's moving speed, direction, and posture changes. Considering that during the actual movement process, the changes in standing pressure and the changes in motion state are interrelated. For example, when the patient's body tilts forward, the motion state data will show a forward acceleration, and at the same time, the standing pressure data will show an increase in the sole pressure. Therefore, in order to comprehensively consider the standing pressure characteristics and three-dimensional motion state characteristics of the patient, the present application further cascades the standing pressure time-series context-associated encoding vector and the motion state time-series context-associated encoding vector to achieve multi-source information complementarity, obtain a motion state multi-modal cascade encoding vector, and thus more comprehensively describe the patient's motion state.

[0059] Specifically, in a specific example of the present application, the classification and recognition unit 152 is configured to: perform fully connected encoding on the motion state multi-modal cascade encoding vector using the fully connected layer of the abnormal behavior detector to obtain a motion state multi-modal cascade fully connected encoding vector; input the motion state multi-modal cascade fully connected encoding vector into the Softmax classification function of the abnormal behavior detector to obtain the probability values of the motion state multi-modal cascade encoding vector belonging to each classification label, where the classification labels include that the patient object's action is abnormal and the patient object's action is not abnormal; and determine the classification label corresponding to the maximum value among the probability values as the detection result. Specifically, by further encoding the motion state multi-modal cascade encoding vector through the fully connected layer of the abnormal behavior detector, more abstract and high-level feature representations can be captured, which helps to improve the model's understanding ability of complex patterns and makes the abnormal behavior detection more accurate. Using the Softmax classification function, a probability value is assigned to each possible classification label (i.e., the patient object's action is abnormal and not abnormal), and the probability distribution reflects the model's evaluation of the confidence of different categories, increasing the transparency and reliability of the decision-making process. Finally, the classification label with the highest probability value is selected as the final output, ensuring the response speed and efficiency of the system.

[0060] In a preferred example of the present application, inputting the motion state multi-modal cascade encoding vector into the abnormal behavior detector based on a classifier to obtain a detection result includes:

[0061] First, determine the eigenvalue mean μ and eigenvalue standard deviation σ corresponding to the motion state multi-modal cascade encoding vector;

[0062] Secondly, subtract the dot product vector of the eigenvalue mean from the multi-modal cascade coding vector of the motion state and multiply it by the eigenvalue standard deviation to obtain the first multi-modal cascade coding fairness target vector of the motion state, which is expressed by the formula:

[0063]

[0064] where μ and σ respectively represent the eigenvalue mean and eigenvalue standard deviation corresponding to the multi-modal cascade coding vector of the motion state, V represents the multi-modal cascade coding vector of the motion state, denotes subtraction by position, ⊙ denotes multiplication by position, and V 1 represents the first multi-modal cascade coding fairness target vector of the motion state;

[0065] Then, subtract the dot product vector of the eigenvalue standard deviation from the multi-modal cascade coding vector of the motion state and multiply it by the eigenvalue mean to obtain the second multi-modal cascade coding fairness target vector of the motion state, which is expressed by the formula:

[0066]

[0067] where V 2 represents the second multi-modal cascade coding fairness target vector of the motion state;

[0068] Next, multiply the element-wise reciprocal of the second multi-modal cascade coding fairness target vector by the first multi-modal cascade coding fairness target vector, and then take the element-wise base-2 logarithm to obtain the multi-modal cascade coding information correction vector of the motion state, which is expressed by the formula:

[0069]

[0070] where, denotes the element-wise reciprocal of the vector, and V in represents the multi-modal cascade coding information correction vector of the motion state;

[0071] Then, take the square root of the quotient of dividing the eigenvalue mean μ by the eigenvalue standard deviation σ, multiply it by the weight hyperparameter, and then add it to the multi-modal cascade coding information correction vector of the motion state to obtain the optimized multi-modal cascade coding vector of the motion state, which is expressed by the formula:

[0072]

[0073] where δ represents the weight hyperparameter, denotes addition by position, and V′ represents the optimized multi-modal cascade coding vector of the motion state;

[0074] Finally, input the optimized multi-modal cascaded coding vector of the motion state into an anomaly behavior detector based on a classifier to obtain a detection result.

[0075] Here, since the standing pressure time-series context correlation coding vector and the motion state time-series context correlation coding vector respectively represent the forward time-series context semantic correlation features of standing pressure and the time-series context significant transmission coding features of the motion state, when they are cascaded to obtain the multi-modal cascaded coding vector of the motion state, the heterogeneity of the feature distribution manifolds of the standing pressure time-series context correlation coding vector and the motion state time-series context correlation coding vector will cause semantic cascading fairness differences in the multi-modal cascaded coding vector of the motion state, thereby affecting the response inclusiveness of the coding feature distribution of the multi-modal cascaded coding vector of the motion state and reducing the accuracy of the detection result obtained by inputting it into the anomaly behavior detector based on a classifier.

[0076] Therefore, considering the fairness differences in the attribute levels of the data population corresponding to the time-series cascaded features of the multi-modal cascaded coding vector of the motion state, in order to improve the response inclusiveness under the feature distribution diversity of the multi-modal cascaded coding vector of the motion state, by using the cross-probability value constraint based on the multi-modal cascaded coding vector of the motion state as an interactive fairness objective representation, the group feature information interaction propagation of the multi-modal cascaded coding vector of the motion state is corrected, and the unified statistical feature response interaction based on the multi-modal cascaded coding vector of the motion state is used as the feature distribution multi-level fairness objective bias to achieve a robust unified representation of the distribution fairness of the multi-modal cascaded coding vector of the motion state, forming a fair cooperation paradigm under the feature distribution framework of the multi-modal cascaded coding vector of the motion state and improving the accuracy of the detection result obtained by inputting it into the anomaly behavior detector based on a classifier.

[0077] In the above-mentioned auxiliary walking nursing device for neurosurgical care, the safety locking module 160 is used to generate a warning prompt and activate the built-in safety mechanism to lock the position in response to the detection result that there is an abnormality in the patient object's action. It should be understood that the safety locking module contains a series of preset safety strategies and emergency response procedures to minimize accidental risks.

[0078] Specifically, once an abnormality is detected, the alarm device will be immediately activated to send out an audible and visual alarm signal to alert the surrounding people. At the same time, the alarm information will be synchronously sent to the mobile device in the hands of medical staff or the central monitoring platform through the wireless communication interface to ensure that relevant personnel can obtain the notice in the first time. This multi-channel information transmission method greatly shortens the response time and improves the processing efficiency in case of emergency.

[0079] Meanwhile, the assisted walking care device itself will also make rapid adjustments. After the controller installed on the support column receives the abnormal detection result, it will immediately execute the locking instruction. The controller is equipped with a mechanical locking mechanism inside, which can fix the height of the armrest instantly and prevent further movement. This process relies on the pre-set spring force and ratchet structure, and under the coordination of the electronic control unit, the locking action is completed through the precise fit in the sliding sleeve operation cavity. Since the whole process requires no manual intervention, the device can be stabilized in a very short time, avoiding secondary injuries caused by sudden movement changes.

[0080] In addition to physical locking, there is also an electronic control system involved to jointly build a comprehensive safety protection net. For example, the roller part is integrated with an intelligent braking system, which can automatically apply braking force when detecting abnormalities, so that the bottom of the leg stops rolling. The braking system works on the principle of electromagnetic induction. When the controller issues a braking command, the electromagnetic coil is energized to generate a magnetic field, which acts on the metal sheet on the roller shaft, thus realizing non-contact braking. This method not only has a rapid response, but also does not cause wear to the roller, extending its service life while enhancing safety.

[0081] Considering various complex situations that may occur in actual applications, the system reserves a certain degree of manual operation space. Even in the automatic locking state, medical staff are still allowed to manually unlock according to specific situations. For this reason, an emergency unlock button is set on one side of the support column. After pressing the button, the restrictions of the electronic control system can be temporarily lifted, providing the possibility for flexible handling in special situations. In addition, aiming at the possible psychological tension problems of long-term bedridden patients when using the assisted walking device for the first time, the system is specially designed with a progressive locking function. That is, when a slight abnormality is initially detected, it does not immediately lock completely, but first slows down the moving speed, giving enough time for the patient to adapt until it is confirmed that full locking is needed.

[0082] To ensure the effective implementation of all these safety measures, the system conducts self-checks and maintenance regularly. The built-in self-check program will automatically run before each startup to test whether the working status of each component is normal. If a problem is found, a fault code will be displayed on the display screen, and corresponding solutions will be recommended. For some key components, such as sensors, controllers, etc., periodic calibration operations will also be arranged to ensure high precision can still be maintained after long-term use.

[0083] In summary, the auxiliary walking care device for neurosurgical care based on the embodiments of the present application is elucidated. It uses an inertial measurement unit (IMU) and a pressure sensor to collect the three-dimensional motion data and standing pressure data of a patient object respectively, and introduces a neural network model based on deep learning to perform data analysis on the three-dimensional motion data and standing pressure data of the patient object, so as to extract the temporal correlation change characteristics of the patient object's motion state and the dynamic change trend of the standing pressure. Furthermore, by performing fusion analysis on the two, intelligent recognition of the abnormal actions of the patient object can be achieved, and an early warning prompt can be automatically generated and the built-in safety mechanism can be activated to lock the position when an abnormal action occurs. In this way, active safety protection for the patient object can be realized, effectively improving the safety and intelligence level of the auxiliary walking care device, and thus contributing to improving the safety and efficiency of rehabilitation training.

[0084] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, benefits, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, benefits, effects, etc. are essential for each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitating understanding, rather than limitations. The above details do not limit the present invention to necessarily adopt the above specific details to be implemented.

[0085] In the above embodiments, the descriptions of each embodiment have their own emphases. For parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0086] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to cover all changes falling within the meaning and scope of the equivalent elements of the claims in the present invention. Any associated drawing marks in the claims should not be regarded as limiting the claimed rights.

[0087] In addition, it is obvious that the term "including" does not exclude other units or steps, and the singular form does not exclude the plural form. A plurality of units stated in the system claims can also be implemented by a single unit through software or hardware.

[0088] Finally, it should be noted that the above description has been given for purposes of illustration and description. In addition, the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An auxiliary walking nursing device for neurosurgery nursing, comprising two groups of support mechanisms, each group of support mechanisms respectively comprises a support column, a bracket and two legs, the bracket is fixedly connected to the lower end of the support column, the two legs are fixedly connected to the lower end of the bracket, and the bottom of the two legs are both equipped with rollers; the brackets in the two groups of support mechanisms are fixedly connected by a connecting frame, and the upper ends of the two support columns are both fixedly equipped with cross bars; the two sides of the two support columns away from each other are respectively equipped with controllers for adjusting the height of the armrests, and the position of the armrests is fixed by a mechanical locking mechanism, characterized in that: The controller comprises: A motion data acquisition module, used to obtain a time queue of three-dimensional motion data of a patient object acquired by the IMU, and a time queue of standing pressure values ​​acquired by the pressure sensor; A motion state feature extraction module, used for extracting the motion state features of the patient at each time point based on the time queue of the three-dimensional motion data to obtain a time queue of a single-point motion state fully connected embedded coding vector; A motion state temporal context perception module, used for performing temporal context association encoding based on motion state change perception on the time queue of the single-point motion state fully connected embedded coding vector to obtain a motion state temporal context association coding vector; A standing pressure time series pattern feature extraction module, used to extract the standing pressure time series pattern feature of the time queue of the standing pressure value to obtain a standing pressure time series context association encoding vector; A patient action abnormality recognition module, used to determine the detection result of the patient's action state based on the fusion features of the standing pressure temporal context association coding vector and the motion state temporal context association coding vector; The safety locking module is used to generate an early warning prompt and activate a built-in safety mechanism to lock the position in response to the detection result that the patient object's movement is abnormal.

2. The auxiliary walking nursing device for neurosurgery nursing according to claim 1, characterized in that: The three-dimensional motion data includes X-axis acceleration, Y-axis acceleration, Z-axis acceleration, X-axis angular velocity, Y-axis angular velocity and Z-axis angular velocity.

3. The walking assistance nursing device for neurosurgery nursing according to claim 2, characterized in that: The motion state feature extraction module is used to: A single-point motion state embedding encoder based on a fully connected layer is used to perform fully connected embedding encoding on each three-dimensional motion data in the time queue of the three-dimensional motion data to obtain the time queue of the single-point motion state fully connected embedding encoding vector.

4. The auxiliary walking nursing device for neurosurgery nursing according to claim 3, characterized in that: The motion state temporal context perception module comprises: A feature transfer significance measurement unit is used to perform feature transfer significance measurement based on neighborhood motion state change perception on each single-point motion state fully connected embedded coding vector in the time queue of the single-point motion state fully connected embedded coding vector to obtain a time queue of single-point motion state feature transfer significance factors; A context aggregation coding unit is used to perform feature attention modulation context aggregation coding on the time queue of the single-point motion state fully connected embedded coding vector based on the time queue of the single-point motion state feature transmission saliency factor to obtain the motion state temporal context association coding vector.

5. The walking assistance nursing device for neurosurgery nursing according to claim 4, characterized in that: The feature transfer significance measurement unit is used to: Calculate the feature jump degree of each single-point motion state fully connected embedded coding vector in the time queue of the single-point motion state fully connected embedded coding vector to obtain the time queue of the single-point motion state feature jump degree; Calculating the feature transfer space span of each single-point motion state fully connected embedded coding vector in the time queue of the single-point motion state fully connected embedded coding vector to obtain the time queue of the single-point motion state feature transfer space span; Based on the feature jump degree and feature transfer spatial span of each single-point motion state fully connected embedded coding vector in the time queue of the single-point motion state fully connected embedded coding vector, the feature transfer significance factor of each single-point motion state fully connected embedded coding vector is calculated to obtain the time queue of the single-point motion state feature transfer significance factor.

6. The walking assistance nursing device for neurosurgery nursing according to claim 5, characterized in that: The context aggregation coding unit is used to: Inputting the time queue of the single-point motion state feature transfer significant factor into a gated transfer unit including a softmax normalization function to obtain a time queue of the single-point motion state feature transfer significant weight; Based on the time queue of the single-point motion state feature that transfers significant weights, the time queue of the single-point motion state fully connected embedded coding vector is weighted aggregated to obtain the motion state temporal context association coding vector.

7. The auxiliary walking nursing device for neurosurgery nursing according to claim 6, characterized in that: The standing pressure temporal pattern feature extraction module is used to: The time queue of the standing pressure value is input into the standing pressure time series pattern feature extractor based on the forward LSTM model to obtain the standing pressure time series context association encoding vector.

8. The walking assistance nursing device for neurosurgery nursing according to claim 7, characterized in that: The patient abnormal movement recognition module comprises: A cascade fusion unit, used for cascading the standing pressure temporal context association coding vector and the motion state temporal context association coding vector to obtain a motion state multimodal cascade coding vector; The classification and identification unit is used to input the motion state multimodal cascade coding vector into a classifier-based abnormal behavior detector to obtain the detection result, and the detection result is used to indicate whether the patient object's action is abnormal.

9. The walking assistance nursing device for neurosurgery nursing according to claim 8, characterized in that: The classification and identification unit is used to: Using the fully connected layer of the abnormal behavior detector to perform fully connected encoding on the motion state multimodal cascade coding vector to obtain a motion state multimodal cascade fully connected coding vector; Inputting the motion state multimodal cascade fully connected encoding vector into the Softmax classification function of the abnormal behavior detector to obtain the probability value of the motion state multimodal cascade encoding vector belonging to each classification label, wherein the classification label includes the patient object movement is abnormal and the patient object movement is not abnormal; The classification label corresponding to the largest probability value among the probability values ​​is determined as the detection result.

Images