Method and device for calculating the probability of an actual fall event after detecting a suspected fall event - Patents.com

Abstract

A method is disclosed for calculating the probability of an actual fall event after detecting a suspected fall event, the method being executed by a processor and including, upon detection of the suspected fall event, obtaining sensor data related to a person over a defined period of time prior to detecting the suspected fall event, generating from the obtained sensor data features related to the person's head movement in a number of stages during the defined period, the features including a vertical velocity of the person's head in each stage and a horizontal displacement of the person's head in each stage, and calculating the probability of an actual fall event based on the generated features.

Description

[Technical Field]

[0001] The present disclosure relates generally to the field of fall detection, and more particularly to methods and devices for detecting a suspected fall event and then calculating the probability of an actual fall event. [Background technology]

[0002] One of the healthcare challenges facing most states and localities in our modern society is caring for a growing senior population, many of whom live alone. A major health risk for vulnerable populations, such as the elderly, infirm, or disabled, is injury from accidental falls, such as in bathrooms. Falls, which may be defined as a sudden, uncontrolled movement of a person's body to the ground or floor, can cause serious health problems or even death if left undetected.

[0003] Therefore, monitoring fall events occurring among vulnerable populations is of great interest to the healthcare industry.Automatic fall detection currently plays a key function in most elderly care solutions.

[0004] Compared with wearable device-based fall detection methods, non-contact fall detection using remote sensors has unique advantages and is therefore becoming mainstream.

[0005] Radar sensors, such as Frequency-Modulated Continuous-Wave Radar sensors, have been employed by many elderly care solutions for many years to enable remote fall detection. In recent years, three-dimensional (3D), time-of-flight (ToF), sensor-based fall detection solutions have emerged.

[0006] In a fall detection system using a ranging sensor such as a 3D ToF sensor, the sensor monitors a space, such as a bathroom, to capture the position and posture of a person within the space in real time. When a person falls and lies on the floor, the posture of the person perceived by the sensor is different from when the person is standing or sitting.

[0007] Unlike other sensors such as radar sensors and accelerometers in wearable devices, 3D ToF sensors do not focus on the falling behavior / process itself (i.e., before the person lies down on the floor) to detect and confirm a fall.

[0008] Because 3D ToF sensor-based fall detection does not rely on fall behavior, which can vary greatly from person to person and situation to situation, to detect a fall event, the 3D ToF sensor-based fall detection system can significantly reduce false negatives (missing detection) and false positives / alarms.

[0009] On the other hand, 3D ToF sensors, especially those with low resolution, can barely distinguish between a fallen person lying down and a person lying down repairing a toilet / drain pipe, or a bulky object that resembles a person, such as a pile of clothes dumped on the floor before being put into the washing machine, which will trigger false alarms if not handled properly. Summary of the Invention [Problem to be solved by the invention]

[0010] In view of the above, it is desirable to be able to calculate the probability of an actual fall event when a suspected fall event is detected by a fall detection system so as to reduce false alarms by fall detection systems, particularly ToF sensor-based fall detection systems. [Means for solving the problem]

[0011] In a first aspect of the present disclosure, there is provided a method for calculating a probability of an actual fall event after detecting a suspected fall event, the method being executed by a processor and comprising: - upon detection of a suspected fall event, obtaining sensor data relating to the person over a defined period prior to detecting the suspected fall event; - generating features related to the movement of the person's head in a number of stages over a defined period from the acquired sensor data, the features including a vertical velocity of the person's head in each stage and a horizontal displacement of the person's head in each stage; - calculating the probability of an actual fall event based on the generated features; A method is presented, including:

[0012] The present disclosure is based on the insight that the probability that a person is experiencing an actual fall event can be determined based on features related to the person's head movement during a defined period of time prior to detecting a suspected fall event, including the vertical velocity of the person's head and the horizontal displacement of the person's head during the falling procedure leading up to the suspected fall event.

[0013] The method is performed after a suspected fall event is detected. Once the suspected fall event is detected, sensor data is obtained about the person over a defined period of time prior to detecting the suspected fall event. Features related to the person's head movement at a number of stages over the defined period are then generated from the obtained sensor data. These features are then used to calculate the probability of an actual fall event.

[0014] The method is particularly advantageous for fall detection systems that determine a fall event based on the fallen state of a person. The method combines the detection of a potentially fallen person, i.e., a person lying down, with reviewing a pre-lying process after detecting the person lying down.

[0015] This provides a more reliable result regarding the probability that a suspected fall event is an actual fall event, reduces unwanted false positives, and subsequent decisions, such as notifying a caregiver, can be made more reliably.

[0016] In one example of the present disclosure, detecting a suspected fall event includes detecting a person-like object lying on the floor based on sensor data.

[0017] As mentioned above, the present method is particularly advantageous for fall detection based on a person's lying state. This is typically achieved by detecting a person-like object lying on the floor based on sensor data. A person-like object may be detected, for example, as a group of reflections from several zones of the sensor that have a human-like shape and are at a short distance from the floor. Such detection can be easily performed by a ranging sensor.

[0018] In one example of the present disclosure, the obtaining step includes obtaining a number of consecutive measurements taken by the sensor prior to detecting the suspected fall event, the number of consecutive measurements beginning before or at the measurement having the person's highest head position.

[0019] This allows a feature about the person's movement throughout the fall process to be generated, obtained or derived from these measurements, resulting in a more reliable probability of detecting an actual fall. The feature about the person's movement throughout the fall process can be generated at any point during each phase, such as the beginning, end or middle of each phase, or can be an average value across each phase.

[0020] In one example of the present disclosure, the generating step comprises: - obtaining a trajectory of the person's head from the highest head position to the lowest head position based on the obtained sensor data; - dividing the obtained trajectory of the person's head into a number of stages based on height; - calculating the vertical velocity of the person's head at each stage and the horizontal displacement of the person's head at each stage; Includes.

[0021] Dividing the fall process into a certain number of stages allows for more accurate determination of the state or movement of the person in the fall process, as the features include the vertical velocity and horizontal displacement of the person's head in each stage. Therefore, the variation of the features can be taken into account, which helps to obtain an actual fall probability that is closer to the real situation.

[0022] In one example of the present disclosure, if the person's highest moving point is above a threshold, the highest head position is determined as the highest moving point, and if the person's head's highest moving point is below the threshold, the lowest head position is the moving point farthest from the person's feet, and the person's feet is the part of the person closest to the floor.

[0023] This division of the person's head trajectory may be height-based as described herein, i.e., the highest and lowest height values ​​of the trajectory are used as endpoints, and the trajectory is divided evenly in height into several stages.

[0024] In another implementation, an absolute highest and lowest point (eg, 0 meters) is set, for example 1.8 meters, and the height between the highest and lowest points is divided into several steps.

[0025] In one example of the present disclosure, the vertical velocity of the person's head in each stage includes one or more of the vertical velocity of the person's head at any time in each stage, for example, at the beginning of each stage, the end of each stage, or in the middle of each stage, the average vertical velocity of the person's head in each stage, and the weighted vertical velocity of the person's head in each stage.

[0026] Depending on the expected accuracy, different vertical velocities may be calculated and used to calculate the probability of an actual rollover event.

[0027] In one example of the present disclosure, the generating step further includes generating a cross-section of the person at each stage.

[0028] The cross section may be represented as a number of zones of the sensor that a person occupies. This is also a useful feature for determining whether a suspected fall event is real. This can be used in combination with the vertical velocity and horizontal displacement of the person's head to more accurately calculate the probability of a real fall event.

[0029] In one example of the present disclosure, the sensor data is obtained by a ranging sensor, and generating a cross-section of people at each stage includes generating a certain number of zones of the ranging sensor that people occupy at the end of each stage or at any point in each stage.

[0030] In one example of the present disclosure, the calculating step includes inputting the generated features into an artificial intelligence (AI) classifier to obtain a probability of an actual fall event from the AI ​​classifier.

[0031] The principle of determining the probability of an actual fall from the obtained features is that the features, including the vertical velocity, horizontal displacement, and optionally the horizontal cross-sectional size of the person's head at different stages, are together closely related to the person's actual activity type.

[0032] A computer implemented algorithm may be used to infer the actual activity type (fall or not) from the feature values. The algorithm is based on setting thresholds for various features and arriving at a result by jointly considering the comparison results of the different features with their respective thresholds. Any combination of the three features may be used. For simplicity, in one embodiment, only vertical velocity and cross-sectional area are used in each stage. Alternatively, these two features in the final stage may simply be compared with the thresholds.

[0033] For example, if three features related to a person's movement and five levels of height are used, a total of 15 numbers or values ​​will be obtained for the features. These 15 numbers will fluctuate greatly as the person's activity changes. The determination of a fall depends on the absolute value of the numbers, a comparison of the values ​​with thresholds, and their relationship.

[0034] Alternatively, the method may be implemented using an AI classifier that is trained using the tagged dataset and is used to apply threshold criteria to the feature numbers or values ​​to produce a result. This provides a more efficient way of deriving the actual probability of a fall based on the obtained features and associated values.

[0035] In one example of the present disclosure, different weights are assigned to different generated features.

[0036] It can be considered that different motions, postures and activities of a person are represented by different features of the person. Therefore, when determining the probability of an actual fall event, contributions from different features may be considered differently, which may be achieved by assigning different weights to different features. This also helps to improve the accuracy of the calculation of the probability of an actual fall event.

[0037] In a particular example of the present disclosure, in a later stage, a higher weight is assigned to the vertical velocity of the person's head.

[0038] This is based on the idea that in a real fall, a person's head gains speed throughout the fall process until the head hits the floor, and therefore the vertical velocity of a person's head in the later stages is considered to be a more dominant factor in a real fall and is thus given a higher weight.

[0039] In one embodiment of the present disclosure, the method further comprises confirming that an actual fall has occurred if the probability is higher than a threshold.

[0040] Such a further step can be easily realized by performing a comparison operation, which is simple in terms of implementation and consumes little computational resources.

[0041] In one embodiment of the present disclosure, the sensor data is obtained by a Time of Flight (ToF) sensor.

[0042] ToF sensors, which are readily available on the market, are a good choice for collecting sensor data and can be used to obtain the sensor data required for the methods of the present disclosure.

[0043] A second aspect of the present disclosure provides an electronic device configured to perform the method according to the first aspect of the present disclosure.

[0044] A third aspect of the present disclosure provides a computer program product including a computer-readable storage medium storing instructions that, when executed on at least one processor, cause the at least one processor to perform a method according to the first aspect of the present disclosure.

[0045] The above and other features and advantages of the present disclosure will be best understood from the following description taken in conjunction with the accompanying drawings, in which like reference numerals indicate identical parts or parts that perform the same or equivalent functions or operations. [Brief explanation of the drawings]

[0046] [Figure 1] 1a and 1b show the target perceived by the ToF sensor in a standing and lying position, respectively. [Figure 2] FIG. 2 illustrates a schematic diagram of a fall detection system according to the present disclosure. [Figure 3] FIG. 3 illustrates, in a flow chart type diagram, an embodiment of a method for calculating the probability of an actual fall event after detecting a suspected fall event according to the present disclosure. [Figure 4] 4(a) and 4(b) show a schematic and exemplary view of a person in a space perceived by a ToF sensor and the trajectory of the person's head obtained from ToF sensor data. [Figure 5] FIG. 5 shows a schematic representation of a human head trajectory divided into several stages and exemplary features generated for stage 4. [Figure 6] FIG. 6 shows a schematic diagram of the AI ​​classifier used in this disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0047] Embodiments contemplated by the present disclosure will now be described in more detail with reference to the accompanying drawings. The disclosed subject matter should not be construed as limited to only the embodiments set forth herein. Rather, the illustrated embodiments are provided as examples to convey the scope of the subject matter to those skilled in the art.

[0048] Throughout this description, the terms "target," "user," and "subject" are used interchangeably.

[0049] As discussed in the background section, a ranging sensor such as a 3D ToF sensor used in a fall detection system monitors the space in its field of view to capture the position and orientation of a person within the space in real time.

[0050] Specifically, a ToF sensor measures the distance between the sensor and an object in its field of view by emitting a burst of light, usually in the form of infrared light, and measuring the time it takes for the light to bounce off the object and return to the sensor.

[0051] For a ToF sensor with sufficiently high resolution, the light reflected from different parts of the body can be used by the sensor to build a depth map that can represent the 3D shape of the person.

[0052] In contrast, if the ToF sensor is of low resolution, the sensor may not be able to tell from the depth map constructed from light reflected from the person's body to the sensor whether the detected object is a person or some other object such as furniture.

[0053] However, since the posture of a fallen person perceived by the ToF sensor is different from when the person is standing or sitting, the ToF sensor can distinguish between a person lying on the floor and a person standing or sitting.

[0054] Referring to Figures 1a and 1b, which show targets perceived by a ToF sensor in a standing and lying position, respectively, the pillar 11 shown in Figure 1a represents a standing person from the field of view of a low resolution, e.g., 8x8, 3D ToF sensor, while the pillar 12 shown in Figure 1b represents how a person is perceived by the 3D ToF sensor when the person has collapsed and is lying on the floor.

[0055] Based on the above, a 3D ToF sensor can be used to detect a fall event based on reflections from objects that have a human-like shape and are at a short distance from the floor, which are collectively captured by several zones of the sensor.

[0056] However, such reflections can also be caused by bulky (human-like) objects such as a person intentionally lying down repairing a toilet / drain pipe, or a pile of clothes being dumped on the floor before being put into the washing machine, which will trigger a false alarm if not handled properly.

[0057] This disclosure proposes a solution to distinguish between intentionally lying on the floor and actual falls, which may be detected as suspicious falls by ranging sensors. This method helps reduce false alarms in fall detection systems based on low-resolution ToF sensors.

[0058] 2 shows a schematic diagram of a fall detection system 20 according to the present disclosure. The fall detection system includes a ranging sensor, such as a TOF sensor 21, connected to a processing device 23 via a network 22, such as the internet or a home network. The processing device 23 may include or be communicatively connected to a storage device 24.

[0059] The sensor 21 has an emitter 211 that emits a signal, such as an infrared signal, and a receiver 212 that senses the signal reflected from an object, here a person 25, within a detection range or field of view 26 of the sensor 21. The sensor 21 may also include an integrated processor 213 that controls the operation of the sensor 21 and calculates the time of flight of the emitted signal, i.e., the time between when the emitter 211 emits the signal and when the receiver 212 receives the reflected signal.

[0060] Algorithms for detecting motion or presence within the detection range of sensor 21 based on data collected by sensor 21 may be executed on processor 213 of sensor 21 or on a processing device 23 connected to sensor 21 via network 22. Such algorithms may be stored, for example, in storage device 24 or in an internal storage device (not shown) within sensor 21.

[0061] The sensors 21 of the fall detection system 20 may be mounted, for example, on the ceiling of the bathroom to monitor the motion and posture of people in the bathroom without capturing any privacy information.

[0062] When a person enters a space where the sensor is installed and is detected by the sensor 21 via a presence detection algorithm, the sensor 21 begins to monitor the person's motion and posture, while simultaneously recording its measurements for a predetermined duration or period of time for future use.

[0063] In one implementation, a buffer of, for example, 60 seconds is used to store the data. Each time a new measurement is obtained, the buffer deletes the oldest measurement data (if the buffer is full) and stores the new data.

[0064] In order to reduce the size of the recorded data, the sensor may only record measurements in a zone that covers the person.

[0065] The time period may be, for example, 20 seconds. Such measurements or sensor data allow the present disclosure to verify the fall process leading to a suspected fall event once the suspected fall event is detected.

[0066] The sensors 21 continue to monitor the motion and posture of the person. If the sensors 21 detect a possible or suspected fall of the person by identifying a person-like object lying on the floor, the fall detection system begins executing a method to calculate the probability of an actual fall event to confirm whether the suspected fall event is real.

[0067] FIG. 3 illustrates, in a flow chart type diagram, an embodiment of a method 30 for calculating the probability of an actual fall event after detecting a suspected fall event according to the present disclosure.

[0068] The method may be performed by a processor, such as processor 213 integrated into sensor 21 or processor 23, which may be located locally or remotely in the space in which the sensor is installed. Processor 23 may be included in a computing device such as a computer, a mobile phone, or other computing device having sufficient computing power to perform the method of the present disclosure.

[0069] In step 31, upon detection of a suspected fall event, sensor data is obtained about the person over a defined period of time prior to detecting the suspected fall event.

[0070] This may involve, for example, a remote processing device receiving sensor data from a sensor over a defined period of time, or a processor in the sensor retrieving sensor data stored locally within the sensor.

[0071] In step 32, a signature of the person's head movement over a defined period of time is generated from the acquired sensor data.

[0072] The defined period is the period that includes the fall process leading up to the detected suspected fall event. As will be described below, the fall process is divided into several stages based on height in order to improve the accuracy of the calculation results regarding the probability of an actual fall.

[0073] The generated features include a vertical velocity of the person's head at each stage and a horizontal displacement of the person's head at each stage. The generated features may further include a horizontal cross section of the person at each stage.

[0074] In step 33, the generated features are used to calculate the probability of an actual fall event.

[0075] The above procedure is described in detail below with reference to an example of calculating the probability of an actual fall event based on sensor data acquired by a ToF sensor.

[0076] When a suspected fall event is detected by a ToF sensor, for example, by detecting an object such as a person lying on the floor based on sensor data as described above, ToF sensor measurements or sensor data over a period of, for example, 20 seconds are obtained. Because a real fall process is uncontrollable and sudden, it usually takes a very short time, such as within 1-2 seconds. These obtained measurements must have started before or at the time of the measurement with the person's highest head position.

[0077] From the obtained measurements, the position of the person's head is extracted, and thus a trajectory of the person's head is generated. The head position is determined by using the highest moving point as the top of the head if it is above a threshold (e.g., 1 meter), and using the moving point farthest from the feet if it is below the threshold. The feet are the part of the person's body closest to the floor.

[0078] Then, a head trajectory is generated based on the head position. The trajectory starts when the head is at its highest position and ends when the head is at its lowest position. Furthermore, the trajectory is divided into several stages based on the head height.

[0079] 4(a) and 4(b) show a schematic and exemplary view of a person in a space perceived by a ToF sensor and the trajectory of the person's head obtained from the sensor data.

[0080] Referring to FIG. 4(a), a straight line segment 41 represents a standing person, and a small dot 42 at the top end of the straight line segment 41 represents the person's head.

[0081] When the head positions extracted from each measurement over a defined period are plotted on a 3D floor map representing the space where the fall event occurs, a curve 43 is obtained as shown in Figure 4(b). Curve 43 is the trajectory of the person's head movement during the fall process, starting when the head is at its highest position and ending when the head is at its lowest position. In Figures 4(a) and 4(b), reference numerals 44 and 45 represent other objects present in the space where the fall event occurs, such as a sink or toilet.

[0082] The trajectory is then divided into stages based on height, thereby dividing the process of the head as it moves from the highest position to the lowest position into stages, where height can be an absolute height from the ground or floor, or a relative height in the process of falling.

[0083] At each stage, several features are generated that are highly correlated with the posture and activity of the human body. These features include vertical head velocity, horizontal head displacement, and optionally the horizontal cross section of the person.

[0084] Referring to Figure 5, a person's head trajectory 51, shown by a dashed line and illustrating the movement of the head from the highest position to the lowest position, is divided into five stages. To better understand feature extraction from the measurement data at each stage, three different postures of the person are also shown. Specifically, a person in a standing position 52 is shown in light grey, a person in a crouching position 53 is shown in dark grey, and a person lying down 54 is shown in black.

[0085] The vertical height of the trajectory 51 is divided equally into five stages. Alternatively, the division of the trajectory 51 may be based on a height determined by an absolute highest point, for example 1.8 meters, and an absolute lowest point (for example 0 meters). The height that covers the trajectory 51 is set and divided into several stages, and thus the trajectory 51 is also divided into several stages.

[0086] The measurements or sensor data are used to generate features that are highly correlated with the posture and activity of the human body. Such features include the vertical velocity of the person's head and the horizontal displacement of the person's head at each stage. Horizontal cross sections of the person at each stage may also be generated.

[0087] The above features of stage 4 are illustrated in Figure 5. The horizontal axis of Figure 5 is used to indicate the horizontal direction, and the vertical axis indicates the height direction.

[0088] The vertical velocity of the person's head at each stage may include one or more of the vertical velocity of the person's head at any time during each stage, the average vertical velocity of the person's head during each stage, and the weighted vertical velocity of the person's head during each stage. For purposes of simplifying subsequent calculation procedures, the vertical velocity of the head at the end of each stage, shown as downward black arrow 55 in Figure 5, may be used for convenience.

[0089] The horizontal displacement of the head at each stage refers to the distance the head moves at each stage. In Figure 5, an exemplary horizontal displacement of the head at stage 4 is shown by the double arrow 56 pointing left and right.

[0090] The horizontal cross section of the person at each stage is determined where the person's head is at the end of each stage. The cross section can be determined based on a certain number of zones of the ToF sensor that the person occupies. In Figure 5, the part of the person's body above the horizontal line 58 between stages 4 and 5 is the horizontal cross section of the person at the end of the stage.

[0091] Because the ToF ranging sensor has high resolution in the height direction, these features are generated at different head heights.

[0092] Subsequently, the probability that the detected suspected fall event is a real fall event is calculated based on the features generated above, since these features are closely related to the person's actual activity type.

[0093] Setting the threshold is done during the training process. During inference, features are input into the generated model to calculate probabilities. This calculation involves comparisons between features and thresholds, comparisons between different features, and weighted combinations of features and comparison results.

[0094] Specifically, an inference is made from various feature values ​​to the actual activity type. A mathematical model including thresholds for various features is generated in the training process based on the training data. The inference is made by collectively considering the comparison results of different features with their respective thresholds to arrive at a result. This provides the probability that the detected suspected fall event is an actual fall event.

[0095] Such a procedure can be performed using an artificial intelligence (AI) classifier by inputting the generated features of all stages into the AI ​​classifier. Optionally, different weights may be assigned to different features. The weights are determined in a training process and used for inference. For example, more weight can be given to vertical speed in later stages (e.g., stages 4 and 5). The AI ​​classifier is pre-trained using a tagged dataset.

[0096] 6 shows a schematic diagram of an AI classifier used in this disclosure. The AI ​​classifier can be one of the following algorithms, such as XGBoost, LightGBM, neural network, etc., or a combination of several algorithms.

[0097] The AI ​​classifier chooses flexible ensemble models. The first ensemble model, XGBoost / LightGBM, is based on decision trees, and the second neural network is based on logistic regression. Both are very flexible and can adapt to complex mappings.

[0098] These two different types of basic models provide complementary results, which are weighted and combined to obtain better results. This type of ensemble model is suitable for this disclosure because the sensor data used in this disclosure relates to changes in sensor data over time and spatial locations. The problem solved by this disclosure is not a time-series problem because the problem does not occur regularly and there is no fixed relationship between previous and next temporal states.

[0099] The classifier can be deployed on a local device, such as a lighting fixture with an integrated ToF sensor, or on a cloud server. When the AI ​​classifier is deployed on a cloud server, features are generated on an edge device containing the ToF sensor of the fall detection system and transferred to the server via a network, as shown in Figure 1. These data can be further collected and used to iteratively train the AI ​​model and improve its performance.

[0100] The AI ​​classifier generates and outputs an actual fall probability rate.

[0101] In the test cases performed with the AI ​​classifier, 20 actual falls and 20 intentional lying down cases were collected and randomly distributed into a training set (80%) and a test set (the remaining 20%) to train the XGBoost classifier. The classifier showed no errors for the training set and test set even under different partitions of the training / test set.

[0102] The solution combines a method for detecting a person lying down with a post-detection, pre-lying process verification, which helps reduce false positives in ToF ranging sensor-based fall detection systems.

[0103] The system's computational power requirements are not significant due to the following facts: First, using ToF ranging data, human-like objects on the floor can be easily detected; second, the fall confirmation process, including AI classification, only works if a human-like object is detected.

[0104] The present disclosure is not limited to the examples disclosed above, but can be modified and extended by those skilled in the art beyond the scope of the present disclosure disclosed in the appended claims without the need to apply inventive skills, for use in any data communication, data exchange and data processing environment, system or network.

Claims

1. 1. A method for calculating a probability of an actual fall event after detecting a suspected fall event, the method being executed by a processor and comprising: Upon detecting a suspected fall event, obtaining sensor data about the person over a defined period of time prior to detecting the suspected fall event; generating features related to the person's head movement at a number of stages of the defined time period from the obtained sensor data, the features including a vertical velocity of the person's head at each stage and a horizontal displacement of the person's head at each stage; calculating a probability of an actual fall event based on the generated features; Including, The method, wherein the sensor data obtained by the processor is obtained by a 3D time-of-flight sensor.

2. The method of claim 1 , wherein detecting the suspected fall event includes detecting an object, such as a person, lying on a floor based on sensor data.

3. 3. The method of claim 1 or 2, wherein the obtaining step includes obtaining a number of consecutive measurements taken by a sensor prior to detecting the suspected fall event, the number of consecutive measurements beginning before or at a measurement having a highest head position of the person.

4. The generating step includes: obtaining a trajectory of the person's head from a highest head position to a lowest head position based on the obtained sensor data; dividing the obtained trajectory of the person's head into a number of stages based on height; calculating a vertical velocity of the person's head at each stage and a horizontal displacement of the person's head at each stage; The method of claim 3, comprising:

5. 5. The method of claim 3 or 4, wherein if the highest moving point of the person's head is above a threshold, the highest head position is determined as the highest moving point, and if the highest moving point of the person's head is below the threshold, the lowest head position is the moving point farthest from the person's feet, and the person's feet is the part of the person closest to the floor.

6. 6. The method of claim 1, wherein the vertical velocity of the person's head at each stage comprises a vertical velocity of the person's head at any time during each stage, an average vertical velocity of the person's head at each stage, and a weighted vertical velocity of the person's head at each stage.

7. The method of claim 1 , wherein the generating step comprises generating a cross-section of the person at each stage.

8. 8. The method of claim 7, wherein the sensor data is obtained by a ranging sensor, and generating a cross-section of the person at each stage includes generating a number of zones of the ranging sensor occupied by the person at the end of each stage.

9. 9. The method of claim 1, wherein the calculating step comprises inputting the generated features into an artificial intelligence (AI) classifier to obtain a probability of an actual fall event from the AI ​​classifier.

10. The method of claim 9 , wherein different generated features are assigned different weights.

11. The method of claim 10 , wherein in a later stage a higher weight is assigned to the vertical velocity of the person's head.

12. 12. A method according to any preceding claim, comprising confirming that an actual fall has occurred if the probability is higher than a threshold.

13. An electronic device configured to perform the method of any one of claims 1 to 12.

14. 13. A computer program product comprising a computer readable storage medium having stored thereon instructions which, when executed on at least one processor, cause the at least one processor to perform the method of any one of claims 1 to 12.

Images