TWI767731B - Fall detection system and method - Google Patents

Abstract

A fall detection system includes a radar that generates emitting radio waves and receives reflected radio waves from a person under detection, a data generator that generates a point cloud according to the reflected radio waves, an area determining device that determines a sub-area of a detecting area in which the person under detection lies, and a classifier that determines whether the person under detection falls according to the point cloud. The classifier adaptively processes the point cloud with different methods according to sub-areas as determined by the area determining device respectively to determine whether the person under detection falls.

Description

Fall detection system and method

The present invention relates to a fall detection, and more particularly, to an adaptive fall detection system and method.

Falls in the elderly are a common and serious health problem, and if falls are detected and reported early, immediate medical attention can be provided. The wearable detection device is one of the traditional fall detection mechanisms, and belongs to the contact type detection device. The disadvantage is that it needs to be worn at any time, thus causing inconvenience.

Another traditional fall detection mechanism is to use an image capture device (such as a camera) to continuously capture images, and to determine whether the person to be detected falls by analyzing the image content, which is a non-contact type or a contactless type. A detection device, such as disclosed in US Patent Application No. 17/085,683, entitled "FALL DETECTION AND REPORTING TECHNOLOGY". However, this technique cannot preserve the privacy of the person to be detected, so it is not suitable for some situations, such as bathrooms.

Another traditional fall detection mechanism uses a radar device to transmit radio waves, and analyzes the reflected radio waves to determine whether the person to be detected has fallen, which is also a non-contact detection device, such as disclosed in US Patent Application No. 16/590,725, The title is "DETECTION METHOD, DETECTION DEVICE, TERMINAL AND DETECTION SYSTEM". However, the radio wave emission and reflection of radar devices cannot achieve comprehensive detection, resulting in too low detection accuracy in some areas, which often results in false alarms, waste of manpower and resources, or false alarms, thus missing medical opportunities.

Therefore, it is urgent to propose a novel detection mechanism to improve the shortcomings of traditional fall detection.

In view of the above, one of the objectives of the embodiments of the present invention is to provide a fall detection system and method, which adaptively adopts different methods to process point cloud data according to the sub-region where the person to be detected is located. In this way, the accuracy of fall detection can be greatly improved, and false positives or false negatives can be reduced.

According to an embodiment of the present invention, a fall detection system includes a radar, a data generator, an area determination device, and a classifier. Radar is used to generate transmitted radio waves and receive reflected radio waves from the person to be detected. The data generator generates point cloud data based on the reflected waves. The area determination device determines in which sub-area of the detection area the person to be detected is located according to the point cloud data. The classifier determines whether the person to be detected falls according to the point cloud data. The classifier adapts different methods to process the point cloud data according to the sub-area where the person to be detected is determined by the area determination device, so as to determine whether the person to be detected falls.

The first figure shows a block diagram of a fall detection system 100 according to an embodiment of the present invention, and the second figure shows a flowchart of a fall detection method 200 according to an embodiment of the present invention. Although this embodiment takes the fall detection of elderly care as an example, the present invention can also be applied to other applications, such as fall detection of toddlers.

The fall detection system 100 of this embodiment may include a radar 11 for generating emitting radio waves 111 and receiving reflected radio waves 112 from the person to be detected 10 (step 21 ). In this embodiment, the radar 11 includes a millimeter wave (millimeter wave or mmWave) radar, and the frequency of the radio waves generated by the radar 11 is 30-300 GHz, and the wavelength is about 1-10 mm. In one embodiment, the radar 11 may comprise an ultra wideband (UWB) radar, which is suitable for short range, low power consumption applications.

The fall detection system 100 of this embodiment may include a data generator 12 for generating point cloud data (step 22 ) according to the reflected radio waves 112 , which includes a plurality of three-dimensional data for representing the three-dimensional data of the person to be detected 10 . shape. Generally speaking, a point cloud refers to a collection of data points in space to represent the three-dimensional shape of an object, wherein each data point has three-dimensional coordinates.

The fall detection system 100 of the present embodiment may include an area determination device 13, which determines which sub-area of the detection area the person to be detected 10 is located in according to the point cloud data, for example, determines the to-be-detected area according to the position in the point cloud data set The position of test taker 10.

The third figure illustrates a top view of the detection area 300, wherein 31 represents the main field of view line of the radar 11, and 32 represents the boundary line. By the main line of sight 31 and the boundary line 32 , the detection area 300 can be divided into the following sub-areas: an ordinary zone 301 , a dead zone 302 and a peripheral zone 303 . Generally speaking, the general area 301 refers to the coverage area or the area within the line of sight 31 of the radar 11 , the blind area 302 refers to the area that is too close to the radar 11 to receive the reflected radio wave 112 correctly, and the boundary area 303 refers to the area outside the coverage area of the radar 11 or the line of sight 31 or the area. It is worth noting that the intensity of the radio waves in the general area 301 is much larger than that in the dead zone area 302 or the boundary area 303 (for example, the ratio of the radio wave intensity is greater than an adjustable threshold), so the accuracy of the point cloud data in the general area 301 is much greater than Accuracy of point cloud data in dead zone 302 or border zone 303.

The fall detection system 100 of this embodiment may include a classifier 14, which determines whether the person to be detected 10 falls according to the point cloud data. In one embodiment, the classifier 14 may include a neural network, which extracts feature values from point cloud data through training, so as to determine whether the person to be detected 10 falls. According to one of the features of the present embodiment, the classifier 14 adaptively uses different methods to process the point cloud data according to the sub-area where the to-be-detected person 10 is determined by the area determination device 13 to determine whether the to-be-detected person 10 falls.

In step 23 , the area determination device 13 determines whether the person to be detected 10 is located in the general area 301 . If the determination result of step 23 is affirmative (that is, the to-be-detected person 10 is located in the general area 301 ), the flow proceeds to step 24 .

In step 24, the classifier 14 determines the size and height change rate of the point cloud data, so as to determine whether the person to be detected 10 falls (step 25). For example, when the size of the point cloud data is smaller than the preset first threshold value and the height change rate is greater than the preset second threshold value, it is determined that the person to be detected 10 has fallen. The determination method of steps 24-25 can use conventional methods, such as disclosed in US Patent Application No. 16/590,725, entitled "DETECTION METHOD, DETECTION DEVICE, TERMINAL AND DETECTION METHOD, DETECTION DEVICE, TERMINAL AND DETECTION SYSTEM).

If it is determined in step 25 that the person to be detected 10 has fallen, the process goes to step 26 to issue a warning message; otherwise, the process returns to step 21 .

If the result of the decision in step 23 is negative (that is, the to-be-detected person 10 is located in the blind area 302 or the boundary area 303 ), the process proceeds to step 27 . At step 27, the classifier 14 determines whether the point cloud data (which represents the subject 10 to be detected) has moved within a predetermined period. For example, when the moving distance of the point cloud data within a predetermined period (eg, 60 seconds) is less than a predetermined threshold, it is determined that the point cloud data has not moved. If it is determined in step 27 that the point cloud data has not moved within the preset period, indicating that the person to be detected 10 is likely to fall, the process goes to step 26 to issue a warning message (the message format may be the same or different from that of the person to be detected 10 ). The warning message issued by falling in the general area 301 ); otherwise, the flow returns to step 21 . Since step 27 determines whether the person to be detected 10 (located in the dead zone 302 or the boundary area 303 ) falls according to whether the point cloud data moves The area 301 is avoided to be located in the dead zone area 302 or the boundary area 303 , so as to avoid erroneously determining that the person to be detected 10 leaves the detection area 300 through the entrance and exit 33 as a fall.

According to the above-mentioned embodiment, the classifier 14 adaptively adopts different methods to process the point cloud data according to the sub-region where the person to be detected 10 is located and the accuracy of the point cloud data in the sub-region. In this way, the accuracy of fall detection can be greatly improved, and false positives or false negatives can be reduced.

The above descriptions are only preferred embodiments of the present invention and are not intended to limit the scope of the patent application of the present invention; all other equivalent changes or modifications made without departing from the spirit disclosed in the invention shall be included in the following within the scope of the patent application.

100: Fall Detection System 10: To be detected 11: Radar 111: launch radio waves 112: Reflected radio waves 12: Data Generator 13: Area determination device 14: Classifier 200: Fall detection method 21: Transmit and receive radio waves 22: Generate point cloud data 23: Decide whether to be located in the general area 24: Determine the size and height change rate of the point cloud data 25: Determine whether to fall 26: Issue a warning message 27: Determine whether the point cloud data is moving 300: Detection area 31: Line of sight 32: Boundary Line 33: Entrance and Exit 301: General area 302: Dead Zone 303: Boundary Zone

The first figure shows a block diagram of a fall detection system according to an embodiment of the present invention. The second figure shows a flowchart of a fall detection method according to an embodiment of the present invention. The third figure illustrates a top view of the detection area.

100: Fall Detection System

10: To be detected

11: Radar

111: launch radio waves

112: Reflected radio waves

12: Data Generator

13: Area determination device

14: Classifier

Claims (18)

A fall detection system comprising: a radar for generating transmitted radio waves and receiving reflected radio waves from the person to be detected; a data generator, which generates point cloud data according to the reflected radio waves; an area determination device, which determines which sub-area of the detection area the to-be-detected is located in according to the point cloud data; and a classifier, which determines whether the person to be detected falls according to the point cloud data; The classifier adapts different methods to process the point cloud data according to the sub-area where the person to be detected is determined by the area determination device, so as to determine whether the person to be detected falls. The fall detection system of claim 1, wherein the radar comprises a millimeter wave radar. The fall detection system of claim 1, wherein the detection area is divided into the following sub-areas: The general area, the area within the coverage area or line of sight of the radar; dead zone, an area too close to the radar to receive the reflected wave correctly; and Boundary area, the area outside the coverage area or line of sight of this radar. The fall detection system as claimed in claim 3, wherein the intensity of the radio waves in the general area is far greater than the intensity of the radio waves in the blind area or the border area. The fall detection system of claim 1, wherein the classifier comprises a neural network, and extracts feature values from the point cloud data through training, so as to determine whether the person to be detected falls. The fall detection system of claim 3, wherein if the area determining device determines that the to-be-detected person is located in the general area, the classifier determines the size and height change rate of the point cloud data, so as to determine whether the to-be-detected person is located. fall. The fall detection system of claim 6, wherein when the size of the point cloud data is smaller than the preset first threshold and the height change rate is greater than the preset second threshold, it is determined that the person to be detected has fallen. The fall detection system of claim 3, wherein if the area determining device determines that the person to be detected is not located in the general area, the classifier determines whether the point cloud data moves within a predetermined period. The fall detection system of claim 8, wherein when the moving distance of the point cloud data within a predetermined period is less than a predetermined threshold, it is determined that the person to be detected has fallen. A fall detection method, comprising: Generate transmitted radio waves and receive reflected radio waves from the person to be detected; to generate point cloud data based on the reflected waves; to determine which sub-area of the detection area the subject to be detected is located in based on the point cloud data; and to determine whether the person to be detected falls according to the point cloud data; According to the sub-region where the person to be detected is located, different methods are adapted to process the point cloud data to determine whether the person to be detected falls. The fall detection method of claim 10, wherein the transmission wave is generated by a radar. The fall detection method of claim 11, wherein the detection area is divided into the following sub-areas: The general area, the area within the coverage area or line of sight of the radar; dead zone, an area too close to the radar to receive the reflected wave correctly; and Boundary area, the area outside the coverage area or line of sight of this radar. The fall detection method of claim 12, wherein the intensity of the radio waves in the general area is far greater than the intensity of the radio waves in the blind area or the border area. The fall detection method of claim 10, wherein the point cloud data is processed by a neural network, and feature values are extracted from the point cloud data through training to determine whether the person to be detected falls. The fall detection method of claim 12, wherein if it is determined that the to-be-detected person is located in the general area, the size and height change rate of the point cloud data are determined to determine whether the to-be-detected person falls. The fall detection method of claim 15, wherein when the size of the point cloud data is smaller than a preset first threshold value and the height change rate is greater than a preset second threshold value, it is determined that the person to be detected has fallen. The fall detection method of claim 12, wherein if it is determined that the person to be detected is not located in the general area, it is determined whether the point cloud data moves within a preset period. The fall detection method of claim 17, wherein when the moving distance of the point cloud data within a predetermined period is less than a predetermined threshold, it is determined that the person to be detected has fallen.

Images