Artificial intelligence state recognition babysitting method and system for a stroller

Abstract

The application relates to the field of baby care, and discloses an artificial intelligence state recognition care method and system for a baby carriage, which comprises the following steps: acquiring motion state information of the baby carriage, baby area images of an area where a baby in the baby carriage is located, and environmental information; wherein the baby area images comprise images of the baby and artificial care behaviors of the baby by the outside world, and the environmental information comprises temperature information and humidity information of the area where the baby in the baby carriage is located; through a baby carriage monitoring model based on space-time attention, alarm information when the baby in the baby carriage is monitored is determined according to the motion state information of the baby carriage, the baby area images and the environmental information, and the baby carriage is controlled to send prompt information corresponding to the alarm information. The application can accurately recognize the state of the baby in an outdoor scene, and improves the care effect of the baby.

Description

Technical Field

[0001] This application relates to the field of infant care technology, and more specifically, to an artificial intelligence-based status recognition care method and system for strollers. Background Technology

[0002] Current AI-based infant status recognition and care methods for outdoor scenarios still face significant technical limitations. Firstly, strollers used outdoors are easily affected by human factors; for example, adults actively soothing the infant may lead to false alarms regarding the infant's condition, impacting the accuracy of monitoring. Secondly, in complex outdoor environments, image information is easily affected by strong light, dynamic shadows, or obstructions (such as awnings or blankets), causing a sharp drop in facial expression recognition accuracy. Furthermore, infrared thermal imaging can have temperature detection errors exceeding ±0.5℃ in high-temperature or high-humidity environments. Additionally, inertial measurement units struggle to distinguish between an infant's spontaneous rolling over and passive displacement caused by vehicle bumps. For instance, in high-frequency vibration scenarios such as cobblestone roads, there is a risk of underreporting infant suffocation or the risk of falling.

[0003] Patent application CN118136249A (application number: CN202410247936.8) discloses a hypergraph neural network-based infant health care system, including a hypergraph neural network computing module, an infant emotional information set feedback module, a smart terminal control module, and an infant emotional and psychological feedback comparison table module. The infant emotional information set feedback module collects infant facial images and audio information over a preset time period and transmits them to the hypergraph neural network computing module. This module uses a hypergraph neural network algorithm to obtain a baseline reference body movement amplitude for a specific health care type during a specific time period. By comparing the baseline reference with the real-time health care body movement amplitude, it determines whether to notify the caregiver. When necessary, it calculates the limb positioning posture based on facial images and audio information, plots a real-time limb change amplitude curve, and compares it with a preset limb change amplitude curve to determine the infant's health care needs for notification. Due to the above reasons, the infant state recognition care method in patent application CN118136249A cannot accurately identify the infant's state in outdoor scenarios, resulting in poor infant care effectiveness. Summary of the Invention

[0004] The purpose of this application is to provide an artificial intelligence-based status recognition and care method and system for strollers, which solves the technical problem that existing infant status recognition and care methods cannot accurately identify the infant's status in outdoor scenarios, and achieves the technical effect of accurately identifying the infant's status in outdoor scenarios.

[0005] This application provides an AI-powered status recognition and care method for strollers. The method includes: acquiring the stroller's motion status information, an image of the infant's area within the stroller, and environmental information; wherein the infant's area image includes images of the infant and external human care behaviors towards the infant, and the environmental information includes temperature and humidity information of the area where the infant is located within the stroller; using a stroller monitoring model based on spatiotemporal attention, determining alarm information for monitoring the infant in the stroller based on the stroller's motion status information, the infant's area image, and the environmental information, and controlling the stroller to issue corresponding prompt information; wherein the alarm information includes infant status abnormality alarm information, carelessness alarm information, motion safety alarm information, and infant environment safety alarm information, the infant status abnormality alarm information is used to indicate an abnormal infant status, the carelessness alarm information is used to prompt active human care for the infant, the motion safety alarm information is used to indicate an abnormal stroller motion status, and the infant environment safety alarm information is used to indicate an abnormal environment inside the stroller.

[0006] In one possible implementation, the method further includes: determining the stroller's body frequency information and the baby's movement frequency information based on the stroller's motion state information using a frequency decomposition unit; wherein the stroller's body frequency information includes 5-30Hz stroller vibration frequency information, and the baby's movement frequency information includes 0.5-2Hz low-frequency movement state information; and determining alarm information for monitoring the baby in the stroller based on a spatiotemporal attention-based stroller monitoring model, according to the stroller's body frequency information, the baby's movement frequency information, the baby's area image, and environmental information.

[0007] In another possible implementation, the method further includes: using a skeleton key point tracking unit to determine infant limb activity information and adult arm movement information based on an infant region image; and using a stroller monitoring model based on spatiotemporal attention to determine alarm information when monitoring the infant in the stroller, based on stroller frequency information, infant movement frequency information, infant limb activity information, adult arm movement information, and environmental information.

[0008] In another possible implementation, the method further includes: determining the action confidence of an adult soothing action based on the temporal information of the adult's arm movements using a soothing action detection unit; the action confidence represents the accuracy of the adult soothing action; when the action confidence of the adult soothing action is greater than or equal to a preset action confidence, determining an alarm message for monitoring the infant in the stroller using a stroller monitoring model based on spatiotemporal attention, based on stroller frequency information, infant movement frequency information, infant limb activity information, adult arm movement information, and environmental information; when the action confidence of the adult soothing action is less than the preset action confidence, determining an alarm message for monitoring the infant in the stroller using a stroller monitoring model based on spatiotemporal attention, based on stroller frequency information, infant movement frequency information, infant limb activity information, and environmental information.

[0009] In another possible implementation, the method further includes: using an occlusion detection unit, determining the infant occlusion area ratio and strong light confidence based on the infant region image, where the infant occlusion area ratio represents the size of the occluded area of ​​the infant, and the strong light confidence represents the confidence of the infant image under the influence of strong light; when the infant occlusion area ratio is greater than or equal to a preset infant occlusion area ratio, and the strong light confidence is less than a preset strong light confidence, using a stroller monitoring model based on spatiotemporal attention, determining alarm information for monitoring the infant in the stroller based on stroller frequency information, infant movement frequency information, adult arm movement information, and environmental information; when the infant occlusion area ratio is less than a preset infant occlusion area ratio, and the strong light confidence is greater than or equal to a preset strong light confidence, and when the confidence of the adult soothing action is greater than or equal to a preset action confidence, using a stroller monitoring model based on spatiotemporal attention, determining alarm information for monitoring the infant in the stroller based on stroller frequency information, infant movement frequency information, infant limb activity information, adult arm movement information, and environmental information.

[0010] In another possible implementation, the method further includes: when the infant occlusion area ratio is greater than or equal to a preset infant occlusion area ratio during the first monitoring time period, acquiring adult arm movement information during the first monitoring time period; when the alarm information during the first monitoring time period is a care-absence alarm information, acquiring the action confidence of the adult soothing action of the adult arm movement information during the first monitoring time period; when the action confidence is greater than or equal to a preset action confidence, not issuing the prompt information corresponding to the care-absence alarm information; when the action confidence is less than the preset action confidence, issuing the prompt information corresponding to the care-absence alarm information.

[0011] In another possible implementation, the method further includes: acquiring medical reference information for movement corresponding to the infant's age information and limb movement information; determining the infant movement label corresponding to the infant's limb movement information based on the infant's age information and medical reference information through an infant movement judgment unit; when the infant's occlusion area ratio is less than a preset infant occlusion area ratio, and the strong light confidence level is greater than or equal to a preset strong light confidence level, determining the alarm information for monitoring the infant in the stroller based on a stroller monitoring model based on spatiotemporal attention, according to stroller frequency information, infant movement frequency information, infant limb movement information, infant movement label, adult arm movement information, and environmental information.

[0012] In another possible implementation, the method further includes: obtaining the number of times the baby environment safety alarm information is issued and the number of times the baby status abnormality alarm information is issued within the first time period; when the number of baby environment safety alarms is greater than or equal to the preset number of baby environment safety alarms and the number of baby status abnormality alarms is greater than or equal to the preset number of baby status abnormality alarms, a caregiver absence alarm information is issued.

[0013] In another possible implementation, the method further includes: obtaining the caregiving absence alarm time interval corresponding to the issuance of the caregiving absence alarm information and the motion safety alarm time interval corresponding to the issuance of the motion safety alarm information within the second time period; issuing the caregiving absence alarm information when the caregiving absence alarm time interval is less than the preset caregiving absence alarm time interval and the motion safety alarm time interval is less than the preset motion safety alarm time interval.

[0014] This application also provides an artificial intelligence status recognition and care system for strollers, including a unit for performing the method described in any of the preceding claims.

[0015] The beneficial effects of the embodiments in this application compared with the prior art are:

[0016] This application provides an AI-powered status recognition and care method for strollers. The method includes: acquiring the stroller's motion status information, an image of the infant's area within the stroller, and environmental information; wherein the infant's area image includes images of the infant and external human care behaviors towards the infant, and the environmental information includes temperature and humidity information of the area where the infant is located within the stroller; using a stroller monitoring model based on spatiotemporal attention, determining alarm information for monitoring the infant in the stroller based on the stroller's motion status information, the infant's area image, and the environmental information, and controlling the stroller to issue corresponding prompt information; wherein the alarm information includes infant status abnormality alarm information, carelessness alarm information, motion safety alarm information, and infant environment safety alarm information, wherein the infant status abnormality alarm information is used to indicate an abnormal infant status, the carelessness alarm information is used to prompt active human care for the infant, the motion safety alarm information is used to indicate an abnormal stroller motion status, and the infant environment safety alarm information is used to indicate an abnormal environment inside the stroller. The AI-powered status recognition and care method for strollers in this application embodiment can automatically monitor the baby's status when the stroller is outdoors based on the stroller's movement status, baby area images, and environmental information, thereby improving the protection of the baby when the stroller is used outdoors. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the first artificial intelligence-based status recognition and care method for strollers provided in this application embodiment;

[0019] Figure 2 A schematic diagram illustrating the workflow of the first artificial intelligence-based status recognition and care method for strollers provided in this application embodiment;

[0020] Figure 3 A schematic diagram illustrating the workflow of a second artificial intelligence-based status recognition and care method for strollers provided in this application embodiment;

[0021] Figure 4 A schematic diagram illustrating the workflow of the third artificial intelligence-based status recognition and care method for strollers provided in this application embodiment;

[0022] Figure 5A flowchart illustrating the second artificial intelligence-based status recognition and care method for strollers provided in this application embodiment;

[0023] Figure 6 A flowchart illustrating the third artificial intelligence-based status recognition and care method for strollers provided in this application embodiment;

[0024] Figure 7 This is a schematic diagram of the logical structure of an artificial intelligence status recognition and care system for a stroller, provided as an embodiment of this application. Detailed Implementation

[0025] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0026] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0027] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0028] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0029] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0030] Existing infant condition recognition and care methods cannot accurately identify infant conditions in outdoor settings, resulting in poor infant care outcomes.

[0031] Based on the above reasons, this application provides an artificial intelligence-based status recognition and care method for strollers. The method includes: acquiring the stroller's motion status information, an image of the infant's area within the stroller, and environmental information; wherein, the infant's area image includes images of the infant and external human care behaviors towards the infant, and the environmental information includes temperature and humidity information of the area where the infant is located within the stroller; using a stroller monitoring model based on spatiotemporal attention, determining alarm information for monitoring the infant in the stroller based on the stroller's motion status information, the infant's area image, and the environmental information, and controlling the stroller to issue corresponding prompt information; wherein, the alarm information includes infant status abnormality alarm information, carelessness alarm information, motion safety alarm information, and infant environment safety alarm information, the infant status abnormality alarm information is used to indicate an abnormal infant status, the carelessness alarm information is used to prompt active human care for the infant, the motion safety alarm information is used to indicate an abnormal stroller motion status, and the infant environment safety alarm information is used to indicate an abnormal environment inside the stroller. The AI-powered status recognition and care method for strollers in this application embodiment can automatically monitor the baby's status when the stroller is outdoors based on the stroller's movement status, baby area images, and environmental information, thereby improving the protection of the baby when the stroller is used outdoors.

[0032] In some scenarios, the artificial intelligence status recognition and care method for strollers according to an embodiment of this application can be applied to strollers, enabling efficient monitoring of the stroller's status when it is used outdoors, thereby improving the stroller's usability.

[0033] In other scenarios, the artificial intelligence status recognition and care method for strollers according to the embodiments of this application can be applied to in-vehicle infant safety seats. It can accurately monitor the status of the stroller during use in the in-vehicle infant safety seat and monitor the care status of the infant in the in-vehicle infant safety seat, thereby improving the effectiveness of the use of the in-vehicle infant safety seat.

[0034] The following describes in detail, with specific examples, an artificial intelligence status recognition and care method for strollers provided in the embodiments of this application.

[0035] Figure 1 A flowchart illustrating the first artificial intelligence-based status recognition and care method for strollers provided in this application embodiment is shown below. Figure 1 As shown, the above method includes S110 to S120, and S110 to S120 will be described in detail below.

[0036] S110. Acquire the movement status information of the stroller, the image of the baby area where the baby is located in the stroller, and environmental information. Among them, the baby area image includes images of the baby and external human care behaviors towards the baby, and the environmental information includes the temperature and humidity information of the area where the baby is located in the stroller.

[0037] In the specific implementation of this method, the motion state information of the stroller, the image of the baby's area within the stroller, and environmental information can be acquired. The motion state information of the stroller can be acquired through accelerometers, gyroscopes, and displacement sensors installed on the stroller frame or wheels. Specifically, the accelerometer is used to collect data on the acceleration changes of the stroller during movement, the gyroscope is used to detect changes in the stroller's tilt angle and direction, and the displacement sensor calculates the real-time speed and trajectory by measuring the number of wheel rotations.

[0038] Optionally, the motion status information may also include collision detection data, such as determining whether an abnormal impact has occurred to the vehicle body using pressure sensors on the vehicle body.

[0039] It should be noted that when the artificial intelligence status recognition and care method for strollers in this application embodiment is applied to a car-mounted infant safety seat, the motion status information can be the motion status information of the vehicle detected by vehicle motion detection. The car-mounted infant safety seat obtains the motion status information by communicating with the vehicle's motion detection component.

[0040] In this implementation, the infant area image includes images of the infant and external human care behaviors towards the infant. The acquisition of the infant area image can be achieved by setting an adjustable camera on the roof or side bracket of the stroller. This camera has a wide-angle shooting function to cover the entire body of the infant and human care behaviors within a one-meter radius around the infant.

[0041] Specifically, the image acquisition process can employ dynamic frame rate adjustment technology, automatically switching to a high frame rate mode when the amplitude of the infant's limb movements exceeds a preset threshold. In this implementation, the images of human caregiving behavior include, but are not limited to, visual information such as the caregiver's hand contact actions, the holding status of feeding utensils, and the removal of objects obscuring the infant's face.

[0042] In this implementation, the environmental information includes temperature and humidity information of the area where the baby is located in the stroller. Environmental information is acquired through a multi-node sensor network distributed throughout the stroller's cabin, where temperature sensors are arranged using a combination of contact and non-contact methods. Specifically, contact temperature sensors can be embedded in the support surface of the baby seat to directly measure the baby's skin temperature; non-contact infrared temperature sensors are installed on the top of the cabin to monitor the ambient air temperature distribution. Humidity sensors can use capacitive sensing elements, with their detection nodes located at the intersection of the bottom and side walls of the cabin, effectively avoiding measurement errors caused by direct liquid splashes.

[0043] Optionally, the environmental information can also integrate ultraviolet intensity detection data, and use photosensitive sensors to determine whether the opening and closing status of the awning is suitable for the current lighting conditions.

[0044] S120. Using a stroller monitoring model based on spatiotemporal attention, based on the stroller's motion status information, the baby's area image, and environmental information, determine alarm information for monitoring the baby in the stroller, and control the stroller to issue corresponding prompts for the alarm information. The alarm information includes alarms for abnormal baby status, lack of care, motion safety, and baby environmental safety. Alarms for abnormal baby status indicate an abnormal baby condition; alarms for lack of care prompt active human supervision of the baby; motion safety alarms indicate abnormal stroller motion; and baby environmental safety alarms indicate an abnormal environment inside the stroller.

[0045] In this implementation, a stroller monitoring model based on spatiotemporal attention can be used to determine alarm information when monitoring the baby in the stroller based on the stroller's motion state information, the baby's area image, and environmental information, and control the stroller to issue corresponding prompt information. The stroller monitoring model based on spatiotemporal attention includes a multimodal data fusion module and a hierarchical early warning decision module.

[0046] Figure 2 A schematic diagram illustrating the workflow of the first artificial intelligence-based status recognition and care method for strollers provided in this application embodiment is shown below. Figure 2As shown, in its specific implementation, the stroller monitoring model combines temporal motion sequence analysis with spatial region feature segmentation through a spatiotemporal attention mechanism. In the temporal dimension, the model extracts optical flow features from consecutive frames of infant limb movements to identify abnormal twitching or prolonged periods of stillness. In the spatial dimension, the infant region is divided into three attention regions: head, torso, and limbs, and the motion feature weights for each region are calculated. Environmental and motion state information are then modeled temporally using a gated loop unit to detect the correlation between temperature and humidity changes and the stroller's motion pattern. Through this spatiotemporal attention-based stroller monitoring model, accurate monitoring of the stroller's and infant's states can be achieved by combining spatiotemporal stroller motion state information, infant region images, and environmental information.

[0047] In this implementation, the alarm information can include alarms indicating abnormal infant condition. These alarms can be derived by analyzing information such as body temperature exceeding a preset safety threshold, duration of continuous crying exceeding a set threshold, abnormal respiratory rate, and persistent dangerous posture. Specifically, when the model detects that the infant's mouth and nose are covered by a foreign object for more than 10 seconds, or that the body temperature rises by 1.5 degrees Celsius within 5 minutes, and there is no human intervention during the same time period, a graded alarm signal containing specific abnormal indicators will be generated.

[0048] In this implementation, the alarm information may include care-absence alarm information. The determination of care-absence alarm information is based on a comprehensive analysis of the frequency and duration of human care behavior. For example, if no soothing action by the caregiver is detected for more than 3 minutes during the baby's crying, or if no bottle-holding action is detected after the feeding interval has expired.

[0049] In this implementation, the alarm information may include motion safety alarm information. The generation mechanism of motion safety alarm information can be detected by the motion state risk assessment model in the stroller monitoring model based on spatiotemporal attention. This model establishes a mapping relationship between motion parameters and safety factors through machine learning algorithms. For example, when the stroller is detected to be traveling at a speed exceeding 2 m / s on a downhill section and with a tilt angle greater than 15 degrees, the model will combine historical driving data to predict the rollover risk level.

[0050] In this implementation, the alarm information may include infant environment safety alarm information. The infant environment safety alarm information can be detected by the temperature and humidity safety range model in the stroller monitoring model based on spatiotemporal attention. When the temperature inside the cabin exceeds 28°C and lasts for 5 minutes, or the relative humidity is below 30%, the infant environment safety alarm information is triggered.

[0051] Optionally, the stroller monitoring model based on spatiotemporal attention can also take into account the rate of change of temperature and humidity, for example, issuing an early warning when an abnormal temperature rise trend of 0.5°C per minute is detected.

[0052] For example, when controlling the stroller to emit a prompt message corresponding to the alarm message, the stroller can be controlled to emit a sound or a prompt light corresponding to the alarm message.

[0053] For example, a stroller monitoring model based on spatiotemporal attention can be an attention model based on temporal and spatial attention using a TS-transformer.

[0054] The beneficial effects of the above implementation method are that by fusing the spatiotemporal features of multimodal sensor data, the accuracy and timeliness of infant status recognition are significantly improved, and the false alarm rate is greatly reduced compared with traditional methods; different types of alarm mechanisms enable caregivers to quickly locate the type of risk, which can shorten the emergency response time, improve the accuracy of abnormal event monitoring, and effectively ensure the safety of the infant in the stroller.

[0055] In some implementations, the above method also includes S130 to S140, which are described in detail below.

[0056] S130. Through the frequency decomposition unit, the stroller body frequency information and the baby's movement frequency information are determined based on the stroller's motion state information. The stroller body frequency information includes stroller body vibration frequency information of 5-30Hz, and the baby's movement frequency information includes low-frequency movement state information of 0.5-2Hz.

[0057] In this implementation, a frequency decomposition unit can determine the stroller's frequency information and the baby's movement frequency information based on the stroller's motion state information. During data processing via the frequency decomposition unit, the vibration signal can be decomposed and its features extracted in multiple dimensions based on the stroller's motion state information. Specifically, signal filtering can decompose the original motion state information into high-frequency components reflecting the stroller's mechanical vibration and low-frequency components reflecting the baby's movement characteristics. The stroller's vibration frequency component characterizes the vibration characteristics caused by road bumps or mechanical transmission during stroller movement, while the low-frequency movement component reflects physiological movement characteristics such as the baby's limb movements and changes in body position.

[0058] During implementation, digital signal processing algorithms can be used to perform spectral analysis on the motion data collected by the accelerometer. For example, after converting the time-domain signal to the frequency domain using wavelet transform or Fourier transform, adaptive threshold segmentation technology can be used to separate the feature information of different frequency bands. For the separated vehicle vibration components, their dominant frequency energy distribution characteristics can be further extracted; for the infant's motion components, their periodicity and amplitude variation patterns can be analyzed to establish a motion state recognition model.

[0059] For example, the vehicle body frequency information may include vehicle body vibration frequency information of 5-30Hz, and the infant movement frequency information may include low-frequency movement state information of 0.5-2Hz.

[0060] S140. Using a stroller monitoring model based on spatiotemporal attention, alarm information is determined when monitoring the baby in the stroller, based on stroller frequency information, baby movement frequency information, baby area image, and environmental information.

[0061] Figure 3 A schematic diagram illustrating the workflow of the second AI-powered status recognition and care method for strollers provided in this application embodiment is shown below. Figure 3 As shown, in this implementation, an alarm message for monitoring an infant in a stroller can be determined using a stroller monitoring model based on spatiotemporal attention, based on stroller frequency information, infant movement frequency information, infant area image, and environmental information.

[0062] In this implementation, a comprehensive analysis using a stroller monitoring model based on spatiotemporal attention can achieve accurate monitoring by fusing multimodal perception data. Specifically, the stroller vibration frequency characteristics and the baby's movement frequency characteristics can be encoded into temporal feature vectors, while visual features are extracted from images of the baby's area captured by the onboard camera using a convolutional neural network. In the environmental information processing module, environmental parameters collected from peripheral devices such as temperature and humidity sensors and noise detection modules can be integrated to construct a multidimensional feature space.

[0063] As an optimization, during the model computation phase, the weights of different modalities can be dynamically allocated through a spatiotemporal attention mechanism. For example, when an abnormal increase in vehicle vibration frequency is detected, the model can automatically increase its attention to the infant's posture image and combine visual features to determine whether a safety restraint failure has occurred. When environmental noise exceeds a threshold, the model can simultaneously increase its sensitivity to monitoring the infant's movement frequency, avoiding misjudgments caused by external interference. This multi-dimensional information fusion mechanism enables the system to accurately identify various risk scenarios such as infant crying, abnormal rolling over, and loosening of safety buckles.

[0064] During training, the stroller monitoring model in the above embodiments can have its performance optimized by constructing a joint training framework. For example, a transfer learning method can be used to jointly fine-tune the pre-trained visual recognition model and the frequency decomposition unit in S130, enabling the spatiotemporal attention mechanism to adaptively learn feature relationships in different scenarios.

[0065] The beneficial effects of the above implementation method are that it achieves accurate differentiation between mechanical vibration and physiological action through frequency domain feature decomposition technology, effectively avoiding false alarms caused by signal aliasing in traditional single sensor monitoring; the multimodal fusion mechanism based on spatiotemporal attention significantly improves the monitoring robustness in complex environments; and by integrating environmental perception information and biometric analysis, a multidimensional safety assessment method is constructed, providing reliable technical support for intelligent monitoring of strollers.

[0066] In some implementations, the above method also includes S150 to S160, which are described in detail below.

[0067] S150: Using the skeleton key point tracking unit, determine the infant's limb movement information and the adult's arm movement information based on the infant's region image.

[0068] In this implementation, the skeleton key point tracking unit can determine the infant's limb activity information and the adult's arm movement information based on the infant's region image, thereby improving the accuracy of monitoring the infant's state based on the infant's limb activity information and the adult's arm movement information.

[0069] Specifically, computer vision algorithms can be used to locate and track key points of an infant's skeleton in real time within an image sequence. For example, pre-trained deep neural network models can be used to extract the coordinates of joints in the head, torso, and limbs. For infant limb movement information, the spatiotemporal variation patterns of key point displacement vectors can be analyzed to identify the amplitude and frequency of movements such as rolling over, kicking, and waving. For adult arm movement information, the trajectory of key points of the adult's upper limbs within the stroller's operating area can be compared to determine whether the adult is performing actions such as picking up the infant, adjusting the seatbelt, or leaving the monitored area.

[0070] In image recognition, multi-frame differential technology combined with optical flow field analysis can be used to improve tracking accuracy. For example, joint motion velocity can be calculated by measuring the changes in key point positions between consecutive image frames, and abnormal jitter noise can be filtered out by combining human kinematic constraints. For infant limb activity information, a motion energy model can be further established, fusing joint angle changes and motion acceleration into a composite feature vector; for adult arm movement information, trajectory clustering analysis can be used to distinguish between routine operating behaviors and abnormal departure actions, such as triggering a status marker when an adult arm key point is detected to be continuously moving away from the stroller control area.

[0071] S160. Using a stroller monitoring model based on spatiotemporal attention, alarm information is determined when monitoring the baby in the stroller, based on stroller frequency information, baby movement frequency information, baby limb activity information, adult arm movement information, and environmental information.

[0072] Figure 4A schematic diagram illustrating the workflow of the third AI-powered status recognition and care method for strollers provided in this application embodiment is shown below. Figure 4 As shown, in this implementation, a stroller monitoring model based on spatiotemporal attention can be used to determine alarm information when monitoring the baby in the stroller, based on stroller frequency information, baby movement frequency information, baby limb activity information, adult arm movement information, and environmental information, thereby achieving accurate monitoring of the baby's status by combining baby limb activity information, adult arm movement information, and environmental information.

[0073] When monitoring infant status, a stroller monitoring model based on spatiotemporal attention can be used to fuse multi-source data, enabling precise decision-making by integrating multi-dimensional features of mechanical vibration, physiological movements, and human operation. Specifically, stroller frequency information can be mapped to vibration intensity temporal features, infant movement frequency information can be encoded into physiological activity frequency domain features, infant limb movement information can be converted into a spatiotemporal matrix of joint motion trajectories, and adult arm movement information can be parsed into a probability distribution of operational behavior categories. Simultaneously, environmental information can be used to construct a multi-dimensional vector input model using sensor data (e.g., ambient light intensity, temperature, and humidity) and acoustic features (e.g., ambient noise decibels).

[0074] During the model computation phase, the weight allocation of different feature dimensions can be dynamically adjusted through a spatiotemporal attention mechanism. For example, when the system detects an abnormally high frequency of infant limb movements, the model can automatically increase the attention weight on the infant's region image and combine it with the trajectory of key skeletal points to determine whether there is a risk of falling. When adult arm movement information indicates that the caregiver is absent, the model can simultaneously increase the sensitivity of monitoring the infant's movement frequency and environmental noise to avoid dangerous situations caused by lack of supervision. This fusion mechanism enables the system to effectively distinguish between normal shaking and dangerous movements. For example, when an infant cries and squirms violently, it can combine information about the presence of an adult to determine whether to trigger an alarm.

[0075] The beneficial effects of the above implementation method are that it enables a fine distinction between infant's autonomous activities and adult's operational behaviors through skeleton key point tracking technology, overcoming the misjudgment problem caused by limb occlusion in traditional image recognition. The multimodal fusion mechanism based on spatiotemporal attention enhances the semantic understanding ability in complex scenes, enabling the system to perform joint reasoning by combining mechanical vibration state, physiological action characteristics, and human operational intentions.

[0076] The beneficial effect of the above implementation method is that by introducing the adult behavior monitoring dimension, a collaborative analysis framework of "human-vehicle-environment" is constructed, which significantly improves the scene adaptability and early warning accuracy of the stroller intelligent monitoring system.

[0077] In some implementations, the above method also includes S170 to S180, which will be explained in detail below.

[0078] S170. Using the soothing action detection unit, the action confidence of the adult soothing action is determined based on the time-series information of the adult's arm movements. The action confidence represents the accuracy of the adult soothing action.

[0079] In this implementation, in order to further reduce the error rate of infant state monitoring, a soothing action detection unit can be used to determine the action confidence of adult soothing actions based on the time-series information of adult arm movements. The action confidence represents the accuracy of adult soothing actions, thus achieving the goal of reducing the error rate of infant state monitoring by combining action confidence.

[0080] When performing motion analysis using a soothing motion detection unit, the effectiveness of soothing behaviors can be evaluated based on temporally varying information about adult arm movements. For example, arm joint motion trajectory features can be extracted using a pre-trained posture recognition model, and the elbow flexion and extension angles, wrist movement paths, and hand opening and closing states in consecutive frames can be temporally modeled.

[0081] For calculating action confidence, the real-time detected limb movement patterns can be matched with features in a standard soothing action library. For example, the degree of matching of action trajectories can be quantified by dynamic time warping algorithm, and finally a confidence score representing the standardization of operation can be output.

[0082] S180. When the confidence level of the adult's soothing action is greater than or equal to the preset action confidence level, an alarm message for monitoring the infant in the stroller is determined using a stroller monitoring model based on spatiotemporal attention, based on stroller frequency information, infant movement frequency information, infant limb activity information, adult arm movement information, and environmental information. When the confidence level of the adult's soothing action is less than the preset action confidence level, an alarm message for monitoring the infant in the stroller is determined using a stroller monitoring model based on spatiotemporal attention, based on stroller frequency information, infant movement frequency information, infant limb activity information, and environmental information.

[0083] After obtaining the action confidence level, the monitoring strategy can be adjusted based on the action confidence level when making hierarchical decisions using a stroller monitoring model based on spatiotemporal attention. When the action confidence level of the adult's soothing action is greater than or equal to the preset action confidence level, the stroller monitoring model based on spatiotemporal attention can determine the alarm information for monitoring the baby in the stroller based on stroller frequency information, baby action frequency information, baby limb activity information, adult arm movement information, and environmental information. This allows for monitoring of the baby's status based on the adult arm movement information with higher action confidence.

[0084] As an example, when the confidence level of an adult's soothing actions is high, the system can use the adult's arm movements as a valid monitoring signal in the overall judgment. For instance, when an infant is crying intermittently, if the model recognizes that an adult is performing a regular patting motion, it may reduce the weight given to abnormal infant movement frequency, thus avoiding false alarms caused by the infant's physiological distress.

[0085] After obtaining the action confidence, when the action confidence of the adult soothing action is less than the preset action confidence, the stroller monitoring model based on spatiotemporal attention determines the alarm information for monitoring the baby in the stroller based on stroller frequency information, baby action frequency information, baby limb activity information and environmental information, so as to realize the monitoring of the baby in the stroller when the action confidence is low.

[0086] As an example, when the confidence level of an action falls below a preset standard, the system can automatically switch to independent monitoring mode. In this mode, the model will exclude potentially invalid adult operation information and focus on analyzing the correlation characteristics between vehicle vibration, infant limb movements, and environmental parameters. For instance, when an adult's arm movements are characterized by disordered swinging or prolonged stillness, the system will enhance monitoring of the infant's breathing rhythm and turning frequency, and combine this with changes in ambient temperature and humidity to determine if there is a risk of suffocation, thereby improving the sensitivity of early warning in scenarios where supervision is lacking.

[0087] As an optimization, detection accuracy can be further improved by constructing a three-dimensional motion feature space. For example, shoulder joint rotation angle, forearm swing amplitude, and hand dwell time can be fused into a composite feature vector, and a classifier can be used to determine whether the action conforms to standard comforting paradigms such as hugging, patting, or shaking. For setting the confidence threshold, an adaptive learning method can be adopted to dynamically optimize the judgment criteria based on historical operation data, enabling the system to adapt to the differences in operating habits of different caregivers.

[0088] The beneficial effect of the above implementation method is that, through the action confidence assessment mechanism, it is possible to effectively distinguish between standardized monitoring operations and ineffective limb behaviors, avoid monitoring interference caused by improper operation by adults, and improve the accuracy of alarm judgment.

[0089] The beneficial effect of the above implementation method is that by adopting a hierarchical information fusion strategy, the system can maintain optimal monitoring performance in both scenarios of effective intervention by the guardian and temporary absence, thereby enhancing its adaptability in complex usage environments.

[0090] Figure 5 This is a flowchart illustrating the second artificial intelligence-based status recognition and care method for strollers provided in an embodiment of this application, as shown below. Figure 5 As shown, the above method also includes S210 to S220, which will be described in detail below.

[0091] S210. Through the occlusion detection unit, the occlusion area ratio and strong light confidence are determined based on the infant region image. The occlusion area ratio represents the size of the occluded area of ​​the infant, and the strong light confidence represents the confidence of the infant image under the influence of strong light.

[0092] In this implementation, the occlusion detection unit can determine the infant occlusion area ratio and the strong light confidence level based on the infant region image. The infant occlusion area ratio represents the confidence level for infant state detection and affects the accuracy of infant state detection under occlusion. The strong light confidence level represents the confidence level for infant state detection under the influence of strong light and affects the confidence level of infant state detection when the infant image is affected by strong light.

[0093] When performing image analysis using an occlusion detection unit, the reliability of the monitoring data can be assessed based on the visibility characteristics of the infant region image. Specifically, image segmentation algorithms can be used to distinguish the infant's outline region from occluding objects at the pixel level, and the proportion of the covered area to the overall infant region can be calculated.

[0094] To determine the confidence level of strong light, we can analyze the characteristics of image brightness distribution, such as statistically analyzing the proportion of the highlight area and the peak offset of the histogram, and use the statistical proportion of the highlight area and the peak offset of the histogram as the confidence level of strong light.

[0095] During implementation, multi-scale feature fusion techniques can be used to improve detection accuracy. For example, convolutional neural networks can be used to extract feature maps at different levels, and edge features and texture features can be weighted and fused through an attention mechanism to accurately identify areas of an infant that are covered by a blanket or obscured by toys.

[0096] For the identification of strong light scenes, an illumination compensation model can be established. By analyzing the distribution pattern of overexposed areas in the image and the changes in reflective intensity, the interference of natural light and the influence of continuous strong light sources can be distinguished.

[0097] S220. When the infant's occlusion area ratio is greater than or equal to a preset infant occlusion area ratio, and the strong light confidence level is less than a preset strong light confidence level, the stroller monitoring model based on spatiotemporal attention determines the alarm information for monitoring the infant in the stroller based on stroller frequency information, infant movement frequency information, adult arm movement information, and environmental information. When the infant's occlusion area ratio is less than a preset infant occlusion area ratio, and the strong light confidence level is greater than or equal to a preset strong light confidence level, and the confidence level of the adult's soothing action is greater than or equal to a preset action confidence level, the stroller monitoring model based on spatiotemporal attention determines the alarm information for monitoring the infant in the stroller based on stroller frequency information, infant movement frequency information, infant limb movement information, adult arm movement information, and environmental information.

[0098] When detecting an infant's condition, if the detected infant's occlusion area is large and the lighting conditions are stable, the system can switch to a non-visual-dependent monitoring mode. If the infant's occlusion area ratio is greater than or equal to the preset infant occlusion area ratio, and the strong light confidence level is less than the preset strong light confidence level, it indicates that the infant's occlusion area is too large or the strong light ratio in the infant's area image is too high. In this case, the infant's area image can be discarded for infant condition detection. Through a stroller monitoring model based on spatiotemporal attention, alarm information is determined when monitoring the infant in the stroller based on stroller frequency information, infant movement frequency information, adult arm movement information, and environmental information.

[0099] When monitoring an infant's condition based on vehicle vibration characteristics, infant movement frequency, adult arm movement, and environmental information, the assessment primarily relies on a comprehensive analysis of vehicle vibration characteristics, infant movement frequency, and adult operational behavior.

[0100] For example, when an infant is covered by a thick blanket, causing visual monitoring to fail, the system can analyze the abnormal vibration frequency of the vehicle (such as violent shaking) and the infant's movement acceleration data (such as rapid kicking), combined with information on whether an adult is present to monitor the infant, to determine whether there is a risk of suffocation and trigger an alarm.

[0101] When detecting an infant's condition, if the infant's occlusion area ratio is less than the preset infant occlusion area ratio, and the strong light confidence level is greater than or equal to the preset strong light confidence level, and the adult's soothing action confidence level is greater than or equal to the preset action confidence level, it indicates that the infant's occlusion area is small or the strong light ratio in the infant's area image is small. In this case, infant condition detection can be performed by combining the infant's area image. Using a stroller monitoring model based on spatiotemporal attention, alarm information for monitoring the infant in the stroller can be determined based on stroller frequency information, infant movement frequency information, infant limb activity information, adult arm movement information, and environmental information.

[0102] When monitoring an infant's condition based on vehicle frequency information, infant movement frequency information, infant limb activity information, adult arm movement information, and environmental information, the system can enable an enhanced data fusion strategy when the degree of obstruction is low, strong light interference is minimal, and the adult's operation is standardized.

[0103] For example, when the degree of occlusion is low and the impact of strong light is minimal, if an adult is detected performing standard soothing actions, the model can combine visual features and accurately monitor the state of the baby in the stroller by using information on the stroller's frequency, the baby's movement frequency, the baby's limb movements, the adult's arm movements, and the environment.

[0104] The beneficial effects of the above implementation method are that, through the dual image quality assessment mechanism, the impact of physical occlusion and optical interference on visual monitoring can be effectively identified, providing a reliable basis for multimodal data fusion; and by adopting a condition-triggered dynamic monitoring strategy, the system can automatically switch to a non-visual sensing-dominated mode when the image quality is substandard, ensuring the continuous monitoring capability of key risk indicators.

[0105] The beneficial effect of the above implementation method is that by introducing the adult operating status as a decision condition, a correlation analysis model of environmental interference and human intervention is established, which maintains the robustness and false alarm suppression capability of the monitoring system in complex usage scenarios.

[0106] In some implementations, the above method also includes S230 to S240, which will be described in detail below.

[0107] S230. When the infant occlusion area ratio is greater than or equal to the preset infant occlusion area ratio during the first monitoring time period, obtain the adult arm movement information during the first monitoring time period.

[0108] When monitoring an infant's condition, if the infant's occlusion area ratio is greater than or equal to the preset infant occlusion area ratio during the first monitoring period, it indicates that the accuracy of infant monitoring through infant images is too low. It is possible to obtain the adult's arm movement information during the first monitoring period and determine whether to issue a prompt message corresponding to the caregiving absence alarm based on the adult's arm movement information.

[0109] When analyzing infant monitoring status, time-series correlation analysis can be used to process monitoring data under occlusion scenarios. When occlusion of the infant area is detected within a continuous monitoring period, the characteristics of adult operation behavior in the corresponding time period can be extracted for cross-validation.

[0110] For example, when the infant's occlusion area ratio is greater than or equal to the preset infant occlusion area ratio during the first monitoring period, and the duration of the infant's occlusion reaches the preset threshold, the system can backtrack the adult's arm movement trajectory data during that period and evaluate the effectiveness of the monitoring behavior through an action pattern matching algorithm.

[0111] S240. When the alarm information within the first monitoring time period is a carelessness alarm information, obtain the action confidence of the adult's arm movement information for comforting actions within the first monitoring time period. When the action confidence is greater than or equal to the preset action confidence, do not issue the prompt information corresponding to the carelessness alarm information; when the action confidence is less than the preset action confidence, issue the prompt information corresponding to the carelessness alarm information.

[0112] When determining whether to issue a prompt corresponding to a carelessness alarm, if the alarm information within the first monitoring time period is a carelessness alarm, the action confidence of the adult's arm movements and soothing actions within the first monitoring time period can be obtained. This allows for determining whether to issue a prompt corresponding to a carelessness alarm based on the action confidence, thus avoiding issuing a carelessness alarm after the existence of soothing actions with high action confidence, and improving the accuracy of triggering prompts corresponding to carelessness alarms.

[0113] When determining whether to issue a prompt message corresponding to a carelessness alarm, if the action confidence is greater than or equal to the preset action confidence, it means that there is already a reassurance action with high action confidence, and no prompt message corresponding to the carelessness alarm is issued. If the action confidence is less than the preset action confidence, it means that there is no reassurance action with high action confidence, and a prompt message corresponding to the carelessness alarm can be issued.

[0114] In this implementation, when the system determines there is a risk of carelessness, the alarm logic is optimized through the aforementioned confidence level verification mechanism. For example, if an adult's arm is detected to be making regular swinging movements during the obstruction period and the confidence level is up to standard, even if the vibration sensor shows abnormal vehicle movement, it can be considered an effective reassurance behavior, thus suppressing false alarms. Conversely, when the movement trajectory shows that the adult's arm is stationary for a long time or has moved out of the monitoring area, and the confidence level score is below the threshold, the system will issue a prompt message corresponding to the carelessness alarm. The device can send the prompt message corresponding to the carelessness alarm via audible and visual warnings and push it to a remote mobile device.

[0115] The beneficial effect of the above implementation method is that by establishing the correlation between occlusion events and monitoring behaviors through temporal correlation analysis, the problem of misjudgment caused by visual monitoring failure is effectively solved, and the monitoring reliability in complex scenarios is improved.

[0116] The beneficial effects of the above implementation method are that the alarm review mechanism realizes the dynamic calibration of risk assessment and operation behavior, reduces the frequency of interfering prompts while ensuring safety, and forms a complete evidence chain of monitoring behavior through multimodal data backtracking analysis, providing multi-dimensional verification basis for alarm decision-making, and significantly improving the accuracy of system alarms and user trust.

[0117] In some implementations, the above method further includes: when the alarm information during the first monitoring period is a care-absence alarm information, obtaining the action confidence of the adult soothing action based on the adult arm movement information during the first monitoring period; when the action confidence is less than the preset action confidence, emitting an emotional interactive soothing sound based on voiceprint cloning to soothe the baby in the stroller.

[0118] In this implementation, when the alarm information during the first monitoring period is a care-absence alarm, the confidence level of the adult's arm movements during the first monitoring period is obtained to determine whether further comforting of the infant is needed based on the confidence level of the movements.

[0119] In normal life, infants generally rely more on their mothers or caregivers for comfort, such as voice comfort. In this implementation, the voiceprint cloning technology-based module in the emotional interaction system can learn and master the voice of the mother or caregiver in advance.

[0120] When monitoring an infant's condition, if the confidence level of the action is lower than the preset confidence level, it indicates that the adult's soothing actions do not meet the requirements for soothing. In this case, the stroller can emit emotionally interactive soothing sounds based on voiceprint cloning to soothe the infant in the stroller, so that the stroller can actively soothe the infant with familiar sounds.

[0121] For example, when a stroller emits emotionally interactive soothing sounds based on voiceprint cloning, it can emit familiar sounds such as "Baby, be good, Mommy's here, don't be afraid..." to soothe the baby.

[0122] The beneficial effect of the above implementation method is that, based on the status monitoring results of the stroller, it is possible to match a scenario suitable for the emotional interaction system, thereby improving the supplementary comfort and care effect for the baby.

[0123] In some implementations, the above method also includes S250 to S260, which will be described in detail below.

[0124] S250. Obtain the medical reference information for movement corresponding to the infant's age information and limb activity information. Using the infant movement judgment unit, determine the infant movement label corresponding to the infant's limb activity information based on the infant's age information and the medical reference information for movement.

[0125] In this implementation, the medical reference information corresponding to the infant's age information and limb activity information can be obtained, and then the infant's condition can be comprehensively judged based on the infant's age information and the medical reference information corresponding to the infant's limb activity information.

[0126] After obtaining the infant's age information and the corresponding medical reference information for movement, the infant movement judgment unit can determine the infant movement label corresponding to the infant's limb movement information based on the infant's age information and the medical reference information for movement.

[0127] For example, when conducting a medical assessment of limb activity based on developmental stage characteristics, a motion standard database corresponding to the developmental stage can be retrieved using the infant's age information input by the user. This database contains typical movement patterns, amplitude ranges, and frequency thresholds for infants of different ages. For infant limb activity information, joint movement trajectories and reference data can be compared in multiple dimensions. For instance, a dynamic time warping algorithm can be used to calculate the similarity of movement morphology, ultimately outputting classification labels characterizing the nature of the movement, such as "physiological stretching," "pathological convulsions," or "emotional tapping."

[0128] For example, for infants aged 0-3 months, the focus is on monitoring neck support and primitive reflexes, while for infants aged 6 months and older, the focus is on analyzing their rolling intentions and ability to maintain a sitting posture.

[0129] S260. When the infant occlusion area ratio is less than the preset infant occlusion area ratio, and the strong light confidence level is greater than or equal to the preset strong light confidence level, the stroller monitoring model based on spatiotemporal attention determines the alarm information for monitoring the infant in the stroller based on stroller frequency information, infant movement frequency information, infant limb activity information, infant movement tags, adult arm movement information, and environmental information.

[0130] When monitoring an infant's condition, if the infant's occlusion area ratio is less than the preset infant occlusion area ratio, and the strong light confidence level is greater than or equal to the preset strong light confidence level, an alarm information for monitoring the infant in the stroller can be determined by a stroller monitoring model based on spatiotemporal attention, based on stroller frequency information, infant movement frequency information, infant limb activity information, infant movement tags, adult arm movement information, and environmental information. This enables monitoring of the infant's condition based on the infant's movement tags.

[0131] In this implementation, when making monitoring decisions by introducing infant action tags, a developmental feature perception alarm mechanism is established to achieve high-precision monitoring. When visual monitoring conditions are met (i.e., less occlusion and adequate lighting), the model can use action tags as an important decision dimension.

[0132] For example, if a 4-month-old infant is detected to have persistent opisthotonus (marked as an abnormality), even if the vehicle vibration frequency is normal, the system will combine ambient temperature data to determine whether the abnormal posture is caused by discomfort, thereby triggering a graded alarm.

[0133] For example, for a 9-month-old infant's attempt to stand independently, the system can reduce the probability of false alarms based on developmental stage characteristics, while strengthening vehicle stability monitoring to prevent the risk of tipping over.

[0134] The beneficial effects of the above implementation method are that by integrating age characteristics and medical knowledge, the clinical guidance value of action recognition is enhanced, and the monitoring system has the ability to adapt to developmental stages; by introducing medical reference labels as a decision dimension, physiological activities and pathological signs are effectively distinguished, and the false alarm rate caused by normal infant developmental behavior is reduced.

[0135] The beneficial effect of the above implementation method is that the multi-dimensional feature space constructed under ideal monitoring conditions enables comprehensive analysis of physiological characteristics, movement patterns and environmental parameters, providing accurate medical decision support for infant safety monitoring.

[0136] Figure 6 This is a flowchart illustrating the third artificial intelligence-based status recognition and care method for strollers provided in this application embodiment. Figure 6 As shown, the above method also includes S310 to S320, which will be described in detail below.

[0137] S310. Obtain the number of times the baby environment safety alarm information was issued and the number of times the baby status abnormality alarm information was issued within the first time period.

[0138] In this implementation, in order to further improve the accuracy of infant status monitoring, when conducting monitoring quality assessment, the overall safety status of the monitoring environment can be analyzed through time-series alarm data. Specifically, the number of infant environment safety alarms and the number of infant status abnormality alarms corresponding to the issuance of infant environment safety alarm information and the issuance of infant status abnormality alarm information can be obtained within the first time period. Then, the infant status can be assessed and monitored based on the number of infant environment safety alarms (y) and the number of infant status abnormality alarms.

[0139] Specifically, a sliding time window can be set to statistically record different types of alarm triggering records within a specified period. Alarms related to infant environmental safety (such as abnormal temperature and humidity, strong light interference) and alarms related to abnormal infant status (such as abnormal breathing, abnormal movement) can be counted independently. When the frequency of the two types of alarms exceeds the preset threshold, their spatiotemporal correlation characteristics can be analyzed, such as detecting the temporal correlation between fluctuations in environmental parameters and changes in physiological indicators.

[0140] During implementation, a dynamic weighting algorithm can be used to optimize alarm thresholds. For example, the sensitivity of environmental safety alarms can be automatically adjusted at different times of day and night, and the counting threshold for light-related alarms can be appropriately reduced during nighttime.

[0141] As an optimization, the statistics of the number of abnormal status alarms can be linked to infant sleep cycle data to distinguish between normal physiological reactions during deep sleep and restless movements during wakefulness, thus avoiding false counting.

[0142] S320. When the number of infant environment safety alarms is greater than or equal to the preset number of infant environment safety alarms, and the number of infant abnormality alarms is greater than or equal to the preset number of infant abnormality alarms, a caregiving absence alarm message is issued.

[0143] In this implementation, when the number of infant environment safety alarms is greater than or equal to the preset number of infant environment safety alarms, and the number of infant status abnormal alarms is greater than or equal to the preset number of infant status abnormal alarms, it indicates that when the environmental safety alarms and status abnormal alarms show clustering characteristics on the time axis, the system can automatically mark them as high-priority events, indicating that the infant may be experiencing a status abnormal alarm due to a poor infant environment or lack of care. At this time, a care-absence alarm message is directly issued.

[0144] As an optimization, when the frequency of dual alarms reaches a certain level, the system can activate a composite risk assessment model to further monitor the infant's condition. For example, if three consecutive alarms indicating excessive temperature and humidity are accompanied by two alarms indicating rapid breathing, the system can determine whether there is a risk of continuous neglect of monitoring by combining the infant's age characteristics (e.g., newborns have weak environmental adaptability) and the caregiver's operation records (e.g., failure to respond to historical alarms in a timely manner).

[0145] The beneficial effects of the above implementation method are that by statistically analyzing alarm frequencies in two dimensions, it is possible to effectively identify the combined risks of environmental factors and infant physiological state, improve the ability to identify monitoring defects in complex scenarios, reduce the probability of misjudgment caused by occasional fluctuations in single alarms, and ensure the objectivity of monitoring quality assessment.

[0146] The beneficial effects of the above implementation method are that the multi-level response system constructed through alarm correlation analysis enables differentiated handling strategies for risk events of different severity levels, which not only avoids excessive interference with the normal monitoring process, but also provides key early warnings for persistent risks, thus comprehensively improving the overall protection effectiveness of the intelligent monitoring system.

[0147] In some implementations, the above method also includes S330 to S340, which will be described in detail below.

[0148] S330. Obtain the time interval between the caregiving absence alarm and the time interval between the movement safety alarm within the second time period.

[0149] In this implementation, when evaluating the effectiveness of monitoring, the operating status of the monitoring system can be analyzed through a time-series alarm mode. Specifically, the time intervals between the care-absence alarm and the movement safety alarm within the second time period can be obtained. The care-absence alarm time interval reflects the caregiver's effectiveness in monitoring the infant's condition, while the movement safety alarm time interval reflects the caregiver's control over the stroller. By combining the care-absence alarm time interval and the movement safety alarm time interval, the protection effect for the infant can be improved.

[0150] For example, a scrolling monitoring window can be set to statistically analyze the time interval characteristics of adjacent alarm events, and time series databases can be established for care-absence alarms (such as long-term absence reminders) and sports safety alarms (such as severe turbulence warnings).

[0151] For example, the calculation of alarm time intervals can be performed using the event stamp difference statistical method to record the density of similar alarm events and the correlation characteristics of cross-alarm events.

[0152] S340. When the time interval for the caregiving absence alarm is less than the preset time interval for the caregiving absence alarm, and the time interval for the motion safety alarm is less than the preset time interval for the motion safety alarm, a caregiving absence alarm message is issued.

[0153] After obtaining the caregiving absence alarm interval and the movement safety alarm interval, if the caregiving absence alarm interval is less than the preset caregiving absence alarm interval and the movement safety alarm interval is less than the preset movement safety alarm interval, it indicates that the caregiving absence alarms and movement safety alarms corresponding to the infant's state are too frequent. The system can activate the enhanced alarm strategy, and at this time, caregiving absence alarm information can be issued.

[0154] As an optimization, the interval threshold can also be dynamically updated using a sliding time window mechanism. For example, the preset interval threshold for motion safety alarms can be automatically shortened when the stroller is in motion to adapt to changes in road conditions during the journey.

[0155] As an optimization, the judgment of the interval for caregiving absence alarms can be personalized by incorporating the guardian's identity information. For example, the response time window after the first alarm can be appropriately extended for new parents. When two types of alarms are detected to be triggered alternately within a short period of time, the system can activate a composite risk analysis model to assess the combined impact on the stability of the guardianship environment and operational standardization.

[0156] For example, if two caregiver absence reminders and three vehicle overload alarms are triggered consecutively within 10 minutes, a caregiver absence alarm can be issued by combining the infant's physiological data (such as the duration of continuous crying) and environmental parameters (such as the continuous abnormal high temperature). At this time, it is determined that there is a risk of lack of monitoring ability.

[0157] The beneficial effects of the above implementation method are that a dynamic evaluation system for monitoring effectiveness is constructed through dual time interval analysis, which can accurately identify the difference between continuous risk events and occasional alarms; and an adaptive threshold adjustment mechanism is adopted to enable alarm strategies to match the characteristic needs of different usage scenarios and user groups, thereby improving the applicability of the system.

[0158] The beneficial effect of the above implementation method is that, through cross-alarm pattern recognition technology, intelligent prediction of complex risk scenarios is achieved, providing decision support for timely implementation of protective measures.

[0159] This application also provides an artificial intelligence status recognition and care system for strollers, including a unit for performing the method described in any of the preceding claims.

[0160] Figure 7 A schematic diagram of the logical structure of an artificial intelligence status recognition and care system for a stroller provided in this application embodiment is shown below. Figure 7 As shown, the system 1 of this embodiment includes a processing unit 11, a storage unit 12, and a transceiver unit 13. The processing unit 11 is used to process data, the storage unit 12 is used to store data, and the transceiver unit 13 is used to send and receive data. The processing unit 11, the storage unit 12, and the transceiver unit 13 cooperate with each other to implement the above-described method. The beneficial effects of the embodiments of this application have been described in the above-described method and will not be repeated here.

[0161] An AI-powered status recognition and monitoring system for strollers, as described in this application, can be applied to strollers. Simultaneously, the AI-powered status recognition and monitoring method for strollers can be applied to in-vehicle infant safety seats. This allows for accurate monitoring of the stroller's status during use in the in-vehicle infant safety seat and also monitors the infant's monitoring status within the seat, thereby improving the effectiveness of the in-vehicle infant safety seat.

[0162] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0163] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0164] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0165] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0166] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0167] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0168] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0169] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. An artificial intelligence state recognition care method for a stroller, characterized by, The method comprises: acquiring motion state information of the stroller, baby area images of an area where a baby in the stroller is located, and environmental information; the baby area images comprise images of the baby and human care behaviors of the baby in the outside world, and the environmental information comprises temperature information and humidity information of the area where the baby in the stroller is located; determining alarm information when monitoring the baby in the stroller according to the motion state information of the stroller, the baby area images, and the environmental information by using a stroller monitoring model based on spatiotemporal attention, and controlling the stroller to output prompt information corresponding to the alarm information; the stroller monitoring model based on spatiotemporal attention comprises a multi-modal data fusion module and a hierarchical early warning decision module; the stroller monitoring model combines time-dimension motion sequence analysis and spatial-dimension area feature division through a spatiotemporal attention mechanism; in the time dimension, the model extracts optical flow features from continuous frame images of baby limb motions, and identifies abnormal convulsions or still timeout states; in the spatial dimension, the baby area is divided into three attention areas, i.e., a head area, a trunk area, and a limb area, and motion feature weights of the areas are calculated respectively; the environmental information and the motion state information are time-series modeled through a gated recurrent unit, and the correlation between temperature and humidity change trends and the motion mode of the stroller body is detected; the alarm information comprises baby state abnormality alarm information, care absence alarm information, motion safety alarm information, and baby environment safety alarm information; the baby state abnormality alarm information is used to prompt a state abnormality of the baby, the care absence alarm information is used to prompt active human care for the baby, the motion safety alarm information is used to prompt a motion state abnormality of the stroller, and the baby environment safety alarm information is used to prompt an environment abnormality in the stroller.

2. The method of claim 1, wherein, The method further comprises: determining stroller body frequency information and baby motion frequency information according to the motion state information of the stroller through a frequency decomposition unit; the stroller body frequency information comprises stroller body vibration frequency information in a frequency range of 5-30 Hz, and the baby motion frequency information comprises low-frequency motion state information in a frequency range of 0.5-2 Hz; determining alarm information when monitoring the baby in the stroller according to the stroller body frequency information, the baby motion frequency information, the baby area images, and the environmental information by using the stroller monitoring model based on spatiotemporal attention.

3. The method of claim 2, wherein, The method further comprises: determining baby limb activity information and adult arm motion information according to the baby area images through a skeleton key point tracking unit; determining alarm information when monitoring the baby in the stroller according to the stroller body frequency information, the baby motion frequency information, the baby limb activity information, the adult arm motion information, and the environmental information by using the stroller monitoring model based on spatiotemporal attention.

4. The method of claim 3, wherein, The method further comprises: determining a motion confidence of an adult soothing motion according to the adult arm motion information in a time sequence through a soothing motion detection unit; the motion confidence represents a correctness of the adult soothing motion. When the action confidence of the adult soothing action is greater than or equal to the preset action confidence, alarm information for monitoring the baby in the baby carriage is determined by the baby carriage monitoring model based on spatiotemporal attention according to the vehicle body frequency information, the baby action frequency information, the baby limb activity information, the adult arm action information, and the environment information; when the action confidence of the adult soothing action is less than the preset action confidence, the alarm information for monitoring the baby in the baby carriage is determined by the baby carriage monitoring model based on spatiotemporal attention according to the vehicle body frequency information, the baby action frequency information, the baby limb activity information, and the environment information.

5. The method of claim 4, wherein, The method further includes: The method further includes: When the baby shielding area ratio is greater than or equal to the preset baby shielding area ratio and the strong light confidence is less than the preset strong light confidence, alarm information for monitoring the baby in the baby carriage is determined by the baby carriage monitoring model based on spatiotemporal attention according to the vehicle body frequency information, the baby action frequency information, the adult arm action information, and the environment information; when the baby shielding area ratio is less than the preset baby shielding area ratio and the strong light confidence is greater than or equal to the preset strong light confidence, and when the action confidence of the adult soothing action is greater than or equal to the preset action confidence, alarm information for monitoring the baby in the baby carriage is determined by the baby carriage monitoring model based on spatiotemporal attention according to the vehicle body frequency information, the baby action frequency information, the baby limb activity information, the adult arm action information, and the environment information.

6. The method of claim 5, wherein, The method further includes: When the baby shielding area ratio is greater than or equal to the preset baby shielding area ratio in the first monitoring time period, the adult arm action information in the first monitoring time period is obtained; When the alarm information in the first monitoring time period is the lack-of-care alarm information, the action confidence of the adult soothing action of the adult arm action information in the first monitoring time period is obtained; when the action confidence is greater than or equal to the preset action confidence, no prompt information corresponding to the lack-of-care alarm information is issued, and when the action confidence is less than the preset action confidence, prompt information corresponding to the lack-of-care alarm information is issued.

7. The method of claim 6, wherein, The method further includes: The method further includes: When the baby shielding area ratio is less than the preset baby shielding area ratio and the strong light confidence is greater than or equal to the preset strong light confidence, alarm information for monitoring the baby in the baby carriage is determined by the baby carriage monitoring model based on spatiotemporal attention according to the vehicle body frequency information, the baby action frequency information, the baby limb activity information, the baby action label, the adult arm action information, and the environment information.

8. The method of claim 7, wherein, The method further includes: acquire a number of baby environment safety alarm information corresponding to the baby environment safety alarm information issued in the first time period and a number of baby state abnormal alarm information corresponding to the baby state abnormal alarm information issued; when the number of baby environment safety alarm information is greater than or equal to a preset number of baby environment safety alarm information, and the number of baby state abnormal alarm information is greater than or equal to a preset number of baby state abnormal alarm information, issue a missing care alarm information.

9. The method of claim 8, wherein, The method further comprises: acquire a missing care alarm time interval corresponding to the missing care alarm information issued in the second time period and a motion safety alarm time interval corresponding to the motion safety alarm information issued; when the missing care alarm time interval is less than a preset missing care alarm time interval, and the motion safety alarm time interval is less than a preset motion safety alarm time interval, issue a missing care alarm information.

10. An artificial intelligence state recognition care system for a baby carriage, characterized by, The device comprises a unit for executing the method of any one of claims 1 to 9.

Images