WO2022130585A1 - 物体検知システムおよび物体検知方法 - Google Patents
物体検知システムおよび物体検知方法 Download PDFInfo
- Publication number
- WO2022130585A1 WO2022130585A1 PCT/JP2020/047246 JP2020047246W WO2022130585A1 WO 2022130585 A1 WO2022130585 A1 WO 2022130585A1 JP 2020047246 W JP2020047246 W JP 2020047246W WO 2022130585 A1 WO2022130585 A1 WO 2022130585A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- detection
- detection device
- learning model
- capture
- object detection
- Prior art date
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 248
- 238000010801 machine learning Methods 0.000 claims abstract description 20
- 238000000034 method Methods 0.000 claims description 15
- 238000004891 communication Methods 0.000 description 26
- 238000012545 processing Methods 0.000 description 8
- 238000009434 installation Methods 0.000 description 7
- 230000005540 biological transmission Effects 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 238000005259 measurement Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V3/00—Electric or magnetic prospecting or detecting; Measuring magnetic field characteristics of the earth, e.g. declination, deviation
- G01V3/12—Electric or magnetic prospecting or detecting; Measuring magnetic field characteristics of the earth, e.g. declination, deviation operating with electromagnetic waves
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W88/00—Devices specially adapted for wireless communication networks, e.g. terminals, base stations or access point devices
- H04W88/02—Terminal devices
Definitions
- the present invention relates to a technique for detecting an object by using supervised machine learning in an object detection system that detects an object in a communication area from the propagation path information of a radio signal communicated between wireless devices.
- Non-Patent Document 1 a method of using supervised machine learning when detecting an object in a communication area from the propagation path information of a radio signal communicated between wireless devices has been studied (see, for example, Non-Patent Document 1).
- an AP Access Point
- STAtion a reference signal to STA (STAtion) to measure the state of the propagation path.
- STAtion a reference signal to STA (STAtion) to measure the state of the propagation path.
- the propagation path information measured by the STA is transmitted to the AP (see, for example, Non-Patent Document 2).
- the object detection system and the object detection method according to the present invention have high accuracy by installing a plurality of detection devices having different performances in a plurality of places in a system for detecting an object in a communication area from the propagation path information of a radio signal.
- the output of the detection device that performs object detection as teacher data, it is possible to easily generate a learning model for object detection using supervised machine learning without spending time or manpower, and the generated learning model can be generated. It can be used with other detectors.
- the present invention captures the propagation path information between the wireless devices in an object detection system that detects an object between the wireless devices based on the propagation path information between two or more wireless devices, and obtains the propagation path information.
- the first detection device includes a capture device that transmits to a first detection device and a second detection device, and a learning model generation device that generates a learning model for object detection using supervised machine learning. Using the learning model generated by the learning model generator, object detection between the wireless devices is performed, and the second detection device performs object detection with higher accuracy than the first detection device.
- the learning model generation device is characterized in that the learning model used in the first detection device is generated and updated using the determination data including the detection result of the object of the second detection device as the teacher data.
- the present invention is an object detection method for detecting an object between the radio devices based on the propagation path information between two or more radio devices, and captures the propagation path information between the radio devices.
- a capture device that transmits propagation path information to a first detection device and a second detection device, and a learning model generation device that generates a learning model for object detection using supervised machine learning are provided.
- the detection device of the above performs object detection between the radio devices using the learning model generated by the learning model generation device, and the second detection device is an object with higher accuracy than the first detection device.
- the learning model generation device performs detection, and is characterized in that the learning model generation device generates and updates the learning model used in the first detection device using determination data including the detection result of an object of the second detection device as training data. And.
- the object detection system and the object detection method according to the present invention have high accuracy by installing a plurality of detection devices having different performances in a plurality of places in a system for detecting an object in a communication area from the propagation path information of a radio signal.
- the output of the detection device that performs object detection as teacher data, it is possible to easily generate a learning model for object detection using supervised machine learning without spending time or manpower, and the generated learning model can be generated. It can be used with other detectors.
- a detection device that makes a highly accurate judgment usually has a large hardware scale and power consumption, it is difficult to install it in the vicinity of the detection location or to install a large number of devices, but it is a device that makes a highly accurate judgment. Can be solved by installing it in a remote location using a network.
- the object detection system measures propagation path information (referred to as CSI (Channel State Information) in the embodiment) between at least two wireless devices such as a base station device and a terminal device, and measures the measurement results.
- CSI Channel State Information
- It is a system that uses a wireless communication system to communicate and captures CSI communicated between wireless devices to detect an object in the communication area of the wireless communication system.
- the wireless communication system is a wireless LAN system will be described, but the same can be applied to any system that measures the state of the propagation path and performs wireless communication.
- FIG. 1 shows an example of the object detection system 100 according to the embodiment.
- AP101 corresponding to a base station device and STA102 corresponding to a terminal device perform communication corresponding to 802.11ac of the wireless LAN standard.
- AP101 has four antennas (AT (1), AT (2), AT (3) and AT (4)). Although the AP 101 shown in FIG. 1 shows an example of four antennas, it suffices if there are two or more antennas. Further, the STA 102 shall be provided with at least one antenna. Further, although FIG. 1 shows an example of one STA102, the present embodiment can be applied even when MU-MIMO (MultiUserMultipleInputMultipleOutput) transmission is performed between the AP101 and a plurality of STA102s. be.
- MU-MIMO MultiUserMultipleInputMultipleOutput
- AP101 transmits VHT NDP (Very High Throughput Null Data Packet) to STA102 as a reference signal for measuring the state of each propagation path between each antenna of AP101 and STA102. Then, the STA 102 calculates the CSI indicating the state of the propagation path between each antenna of the AP101 and the STA102 from the VHT NDP, stores the result in the frame of the VHT Compressed BeamForming Report, and transmits the result to the AP101. The AP 101 performs transmission beam forming processing and the like based on the CSI received from the STA 102. The signals transmitted and received between AP101 and STA102 will be described later.
- VHT NDP Very High Throughput Null Data Packet
- the object detection system 100 includes a capture device 103 (1), a capture device 103 (2), a detection device 104 (1), a detection device 104 (2), and a learning model generation device 106. Have.
- the capture device 103 (1) and the capture device 103 (2) are installed in the vicinity of the communication area including the communication area between the AP 101 and the STA 102, and monitor the wireless LAN frame communicated between the AP 101 and the STA 102. Capture a specific frame.
- the capture device 103 (1) and the capture device 103 (2) determine the VHT Compressed Beamforming Report frame transmitted from the STA 102 to the AP 101 among the wireless LAN frames communicated between the AP 101 and the STA 102, and determine the frame.
- the frame in which the captured compressed CSI is stored and the information on the reception time of the frame are transmitted to both the detection device 104 (1) and the detection device 104 (2) described later.
- the capture device 103 (1) transmits the captured information to both the detection device 104 (1) and the detection device 104 (2), and the capture device 103 (2) sends the captured information only to the detection device 104 (2). You may send it to. Further, the measurement of CSI and the transmission of the measurement result are periodically performed between AP101 and STA102. In the following description, when the capture device 103 (1) and the capture device 103 (2) are common, the (number) at the end of the code is omitted and the description is described as the capture device 103.
- the detection device 104 (1) operates as a first detection device, and information on a frame in which the compressed CSI transmitted from the capture device 103 is stored and the reception time of the frame (hereinafter, capture information as necessary). (Abbreviated as) and is received.
- the detection device 104 (1) detects an object using the learning model generated by the learning model generation device 106, which will be described later.
- the detection device 104 (1) is installed at the same location as the capture device 103 installed near the communication area or near the capture device 103 that does not take time to transfer data. As described above, the detection device 104 (1) has restrictions on the installation location, is smaller in scale than the detection device 104 (2) described later, and has low performance such as processing capacity.
- the detection device 104 (1) detects an object without using a learning model as in the present embodiment, the detection device 104 (1) only detects the presence or absence of an object in real time based on the capture information received from the capture device 103.
- Simple object detection can only be performed with low accuracy.
- object detection is performed using a learning model generated by machine learning using judgment data including the detection result of the detection device 104 (2) that performs highly accurate object detection as teacher data, so that the object detection is low. Highly accurate object detection is possible even with the high-performance detection device 104 (1).
- the determination data includes, for example, the correspondence between a plurality of conditions such as the magnitude and change of the phase and amplitude of each antenna of the AP101 and the detection result such as the position and movement of the object under each condition, and the like, the detection device 104 (2). ) Is various information analyzed by high-precision and high-performance processing.
- the detection device 104 (2) operates as a second detection device and receives the capture information transmitted from the capture device 103. Then, the detection device 104 (2) detects the object based on the captured information.
- the detection device 104 (2) is installed at a distance via the network 105. For this reason, the detection device 104 (2) has less restrictions such as installation location and power supply, and therefore has higher performance such as processing capacity than the detection device 104 (1), and is large-scale capable of processing a large amount of data at high speed. It is possible to install equipment, and it is possible to perform high-precision and advanced object detection processing and analysis that do not require real-time performance. Therefore, the detection device 104 (2) provides the learning model generation device 106 with determination data including the detection result of the object as the teacher data for the learning model generation device 106, which will be described later, to generate the learning model by supervised machine learning. do.
- the network 105 corresponds to the Internet including a communication device such as a network switch and a dedicated communication network.
- the learning model generation device 106 generates a learning model using the determination data including the object detection result of the detection device 104 (2) that performs highly accurate object detection as teacher data. Then, the learning model generation device 106 transmits the generated learning model to the detection device 104 (1). Each time the learning model generation device 106 generates a new learning model, the newly generated learning model is transmitted to the detection device 104 (1), and the learning model of the detection device 104 (1) is the latest learning model. Update to.
- the capture device 103 captures the CSI transmitted from the STA 102 to the AP 101, and the object in the communication area can be detected based on the capture information.
- a plurality of detection devices having different performances are installed at a plurality of places.
- the learning model generation device 106 can detect the object using supervised machine learning without spending time or manpower.
- the learning model of can be easily generated, and the generated learning model can be used in the detection device 104 (1).
- the detection device 104 (2) that makes a highly accurate determination has a large hardware scale and power consumption, it is difficult to install it in the vicinity of the detection location or to install a large number of devices, but the network 105 is used. It is installed in a remote location, and restrictions such as the installation location can be eliminated.
- the capture device 103 (1) and the capture device 103 (2) capture the propagation path information communicated between the pair of wireless devices of the AP 101 and the STA 102, and the capture device 103 (2) captures the capture information.
- another capture device that captures propagation path information communicated between another radio device having an overlapping communication area and another radio device.
- the propagation path information communicated between other wireless devices is captured by another capture device, and the captured information is transmitted only to the detection device 104 (2), and the detection device 104 (2)
- An object within the same range may be detected by using the capture information received from another capture device and the capture information received from the capture device 103 described above.
- the detection device 104 (2) can perform highly accurate and highly functional object detection based on more information, and generate a learning model using more detailed judgment data including the object detection result as teacher data. It can be provided to the device 106 and can generate a highly accurate learning model.
- the detection device 104 (2) may perform object detection by using only a part of the information of the propagation path information captured by the capture device 103 or another capture device.
- FIG. 2 shows an example of a sequence of radio signals communicated between AP101 and STA102.
- AP101 broadcasts a VHT NDP Announcement frame as a start signal of a sounding protocol for acquiring CSI. Immediately after that, AP101 transmits VHT NDP including data for measuring CSI to the destination STA102.
- VHT is an abbreviation for Very High Throughput
- NDP is an abbreviation for Null Data Packet
- VHT NDP is a frame that does not include communication data.
- the VHT NDP Announcement frame includes the addresses of the AP101 and the destination STA102, and is a frame for notifying the STA102 of the transmission of the VHT NDP in advance.
- the VHT NDP Announcement frame is transmitted from one or more specific antennas, and even when transmitted from two or more antennas, the same data signal is transmitted from each antenna.
- the STA 102 that has received the VHT NDP transmitted from the AP 101 derives the CSI value compressed by the method specified by IEEE802.11ac.
- the STA 102 stores the derived compressed CSI in the VHT Compressed Beamforming Report and transmits it.
- the CSI for each antenna of the AP101 is obtained, but as the number of antennas increases, the amount of CSI information fed back to the AP101 increases. Therefore, the CSI (compressed CSI) selected from all the CSIs is fed back to the AP101 by a method predetermined by the wireless LAN standard.
- the capture device 103 captures the CSI measured for each of the four antennas fed back from the STA 102 to the AP 101.
- the object 150 when the object 150 is moving in the direction of the dotted arrow, it enters the communication area from the AT (4) side of the AP101. Then, when the object 150 passes through to the AT (1) side, the CSI of the AT (4) fluctuates first, and the CSI fluctuates in the order of AT (3), AT (2), and AT (1) in time. In this way, by detecting the fluctuation of the CSI for each antenna, it is possible to detect the intrusion and the moving direction of the object 150.
- the above-mentioned detection method is an example, and not only simple object detection that requires real-time performance such as intrusion detection, but also real-time performance can be achieved by analyzing a large amount of CSI accumulated for a predetermined time. It enables highly accurate and highly functional object detection that is not required.
- the detection device 104 (1) which has restrictions on the installation location, detects a simple object such as intrusion detection with low accuracy because of its low performance, but in the present embodiment, it is highly accurate. Since object detection is performed using a learning model generated by machine learning using judgment data including the detection result of the object of the detection device 104 (2) as teacher data, the object is detected with high accuracy even with the low performance detection device 104 (1). Object detection becomes possible.
- FIG. 3 shows a sequence example of the object detection method according to the present embodiment. The sequence shown in FIG. 3 is executed by each device of FIG.
- step (1) as described in FIG. 2, communication is performed between AP101 and STA102.
- the capture device 103 (1) monitors the wireless LAN frame transmitted / received between the AP 101 and the STA 102.
- step (2) the capture device 103 (1) receives only the VHT Compressed Beamforming Report frame (frame in which the compressed CSI is stored) transmitted from the wireless LAN frame to be monitored from the STA 102 specified in advance to the AP 101. Sort and capture.
- step (3) the capture device 103 (1) transmits the information of the frame in which the captured compressed CSI is stored and the reception time of the frame to the detection device 104 (1).
- the capture device 103 (1) including its own number (device-specific identifier such as a serial number) is transmitted to the detection device 104 (1) and the detection device 104 (2).
- the detection device 104 (1) detects an object in real time from the frame in which the compressed CSI received from the capture device 103 (1) is stored and the reception time information of the frame.
- the detection device 104 (1) performs object detection using the learning model received from the learning model generation device 106 in step (11) described later.
- the detection device 104 (1) performs object detection without the learning model, but steps (1) to (11) are , It is repeated in a short cycle, and there is almost no practical problem.
- the capture device 103 (1) receives information on the frame in which the compressed CSI captured in step (2) is stored and the reception time of the frame in parallel with the process of step (3). It is transmitted to the detection device 104 (2).
- all the information captured by the capture device 103 (1) may be transmitted to the detection device 104 (2), or for example, some information may be thinned out to the detection device 104 (2) once every two times. It may be sent to 2).
- the thinning process may be performed on the detection device 104 (2) side.
- the detection device 104 (2) detects an object from the frame in which the compressed CSI received from the capture device 103 (1) is stored and the reception time information of the frame.
- the object detection algorithm used by the detection device 104 (2) has higher accuracy and higher functionality than the object detection algorithm used by the detection device 104 (1).
- the detection device 104 (2) detects an object within the same range as the detection device 104 (1) by using the propagation path information captured by another capture device (for example, the capture device 103 (2)).
- the other capture device may capture the propagation path information between the other radio devices added to the AP101 and the STA102 (for example, the AP101a and the STA102a). In this case, it is assumed that the range of object detection between other wireless devices and the range of object detection between AP101 and STA102 are the same range or overlapping ranges.
- step (7) the detection device 104 (2) transmits the determination data including the detection result of the object to the learning model generation device 106 as teacher data.
- step (8) the above-mentioned processes from step (1) to step (7) are repeatedly executed.
- the processes from step (1) to step (11) may be repeatedly executed, including the processes from the next step (9) to step (11), or the processes from step (1) to step (7) may be executed repeatedly.
- ) May be performed every time the processing up to) is performed a predetermined number of times, and the processing from step (9) to step (11) may be performed.
- the process of steps (9) to (11) may be performed each time the learning model generation device 106 generates a new learning model.
- step (9) the learning model generation device 106 performs supervised machine learning using the determination data including the detection result received from the detection device 104 (2) as supervised data, and generates a learning model.
- step (10) the learning model generation device 106 transmits the learning model generated in step (9) to the detection device 104 (1) via the network 105.
- the detection device 104 (1) refers to the learning model received from the learning model generation device 106, and refers to the frame in which the compressed CSI received from the capture device 103 (1) is stored and the frame. Object detection is performed by analyzing the information of the reception time of.
- step (1) to step (11) are repeated in parallel for the capture device 103 (2) as well. That is, in FIG. 3, the capture device 103 (1) is replaced with the capture device 103 (2), and the operations from step (1) to step (11) are performed.
- the capture device 103 (2) can capture a frame that cannot be captured by the capture device 103 (1), so that the detection device 104 (2) detects an object based on the information received from the plurality of capture devices 103. be able to.
- the determination data including the detection result of the detection device 104 (2) capable of performing the object detection with higher accuracy and higher functionality than the detection device 104 (1) is obtained. Since the detection device 104 (1) performs object detection by referring to the learning model generated by machine learning by the learning model generation device 106 as teacher data, the detection device 104 (1) performs highly accurate object detection in real time. be able to.
- the data may be biased and an appropriate learning model may not be obtained, but the detection device 104 (2) captures. Since the information acquired by the plurality of capture devices 103 such as the device 103 (2) is used, more accurate teacher data can be obtained, so that the learning model generation device 106 can generate a more appropriate learning model.
- FIG. 4 shows an example of the capture information acquired by the detection device 104 from the capture device 103.
- the captured information including the frame in which the compressed CSI is stored and the information on the reception time of the frame is stored in a storage unit such as an internal memory of the detection device 104.
- the capture information includes the reception time when the capture device 103 captures the frame in which the compressed CSI is stored, the AP101 address, the STA102 address, the capture device 103 number, the CSI (acquired CSI) captured by the capture device 103, and the like. Information.
- the addresses of AP101 and STA102 are acquired as the source address and the destination address of the frame in which the compressed CSI is stored. Further, the number of the capture device 103 is a number unique to the capture device 103, and the number of the capture device 103 is added to the information transmitted to the detection device 104 (1) and the detection device 104 (2). The number of the capture device 103 is used to identify each capture device 103 when a plurality of capture devices 103 are arranged. In the example of FIG. 4, the address of the capture target STA 102 preset in the capture device 103 is 11:22:33:44:55:66.
- the CSI transmitted from the STA 102 at the address 11:22:33:44:55:66 to the AP101 at the address AA: BB: CC: DD: EE: FF at the reception time 14:00:00 is the device.
- the CSIs captured by the capture device 103 (1) whose number is (1) and acquired at this time are, for example, ⁇ 11, ⁇ 21, ....
- the CSI transmitted from the STA 102 at the address 11:22:33:44:55:66 to the AP101 at the address AA: BB: CC: DD: EE: FF at the reception time 14:00:01 is the device number.
- Is captured by the capture device 103 (2) of (1), and the CSI acquired at this time is, for example, ⁇ 11, ⁇ 21, ....
- the capture information captured by the plurality of capture devices 103 in chronological order such as reception time 14:00:02 and reception time 14:00:03 is acquired and stored in the internal memory or the like. ..
- the acquired CSI will be described later.
- the reception time is set to every 1 sec and the CSI is captured every 1 sec, but it may be every 10 msec or 100 msec depending on the specifications of the wireless system. good.
- FIG. 4 the case where one AP101 and one STA102 described in FIG. 1 communicate with each other is shown.
- the address is different for each STA102. For example, if there is an STA 102 with an address of 11:22:33:44:55:66 and an STA102 with an address of 22:33:44:55:66:77, capture information is acquired between each STA102 and AP101. Will be. The same applies when there are a plurality of AP101s.
- FIG. 5 shows an example of a compressed CSI transmitted from the STA 102 to the AP 101.
- the number of transmitting antennas (the number of antennas of AP101) ⁇ the number of receiving antennas (the number of antennas of STA102), the number of compressed CSIs, and an example of compressed CSIs are shown.
- the number of transmitting antennas is 2 or more.
- the CSI between each antenna is measured according to the number of antennas of STA102 and the number of antennas of AP101. Therefore, as the number of antennas increases, the amount of CSI information fed back to the AP101 becomes enormous, so the CSI (compressed CSI) selected from the CSI between all antennas according to the conditions determined by the wireless LAN standard is VHT. It is fed back to AP101 by the Compressed Beamforming Report frame.
- the number of transmitting antennas is 2 and the number of receiving antennas is 1 (described as 2 ⁇ 1)
- the number of compressed CSIs is 2, and the compressed CSIs are ⁇ 11 and ⁇ 21.
- the number of compressed CSI is 2
- the compressed CSI is ⁇ 11, ⁇ 21
- the number of compressed CSI is 4
- the compressed CSI is ⁇ 11.
- ⁇ ij corresponds to the phase information between the transmitting antenna number i (i is an integer of 2 or more) and the receiving antenna number j (j is an integer of 1 or more).
- ⁇ ij corresponds to the amplitude information between the transmitting antenna number i and the receiving antenna number j.
- the compressed CSI is determined according to the combination of the number of transmitting antennas and the number of receiving antennas.
- the number of transmitting antennas is the number of antennas of AP101 (4), and the number of receiving antennas is the number of antennas of STA102 (1).
- a total of eight CSIs of four phase information and four amplitude information for each of the four antennas of AP101 are measured.
- 6 compressed CSIs ( ⁇ 11, ⁇ 21, ⁇ 31, ⁇ 21, ⁇ 31, ⁇ 41) are calculated from the measured 8 CSIs, and the calculated compressed CSIs are calculated. Is transmitted to AP101.
- the compressed CSI shown in FIG. 5 is an example.
- ⁇ 11 indicates the phase difference when the signals transmitted from AT (4) and AT (1) are received by the antenna of STA102.
- ⁇ 21 indicates the phase difference between AT (4) and AT (2)
- ⁇ 31 indicates the phase difference between AT (4) and AT (3).
- ⁇ ij ⁇ [0,2 ⁇ ), where i and j are positive integers.
- ⁇ 21 is a value representing the amplitude ratio when the signals transmitted from AT (1) and AT (2) are received by the antenna of STA102 (tan-1 of the ratio of the absolute value of the amplitude). Value) is shown.
- ⁇ 21 indicates the amplitude ratio of AT (1) and AT (2)
- ⁇ 31 indicates the amplitude ratio of AT (1) and AT (3). It should be noted that ⁇ ij ⁇ [0, ⁇ / 2), where i and j are positive integers.
- the STA 102 measures the CSI based on the reference signal transmitted from the AP 101 and transmits the compressed CSI to the AP 101.
- the capture device 103 captures the compressed CSI transmitted from the STA 102 to the AP 101 and transmits it to the detection device 104 (1) and the detection device 104 (2).
- the positions of the capture device 103 (1) and the capture device 103 (2) are near the middle between the AP 101 and the STA 102, but the positions where the signal from the STA 102 can be received. Anything is fine.
- the number of STA 102s is one in FIG. 1, a plurality of STA 102s may be used.
- the present embodiment may be applied to communication between AP101 and another AP101 (in this case, one AP101 functions as STA102).
- the AP 101 or STA 102 may include the functions of the capture device 103 and the detection device 104 (1).
- the AP 101 or STA 102 having the function of the capture device 103 transmits the capture information to the detection device 104 (2) at a distance, receives the learning model from the learning model generation device 106, and receives the learning model from the detection device 104 (1).
- AP101 or STA102 having a function performs object detection.
- the capture device 103 captures the wireless frame in which the compressed CSI transmitted from the STA 102 to the AP 101 is stored, and the communication area is based on the capture information. It is possible to detect the object inside.
- a plurality of detection devices having different performances in the example of FIG. 1, the detection devices 104 (1) and the detection devices 104 (2) are installed at a plurality of places.
- the learning model generation device 106 uses supervised machine learning without spending time or manpower. A learning model for object detection can be easily generated, and the generated learning model can be used in the detection device 104 (1).
- the detection device 104 (2) that performs highly accurate determination has a large hardware scale and power consumption, it is difficult to install it in the vicinity of the detection location or to install a large number of devices. , It is installed in a remote place by using the network 105, and it is possible to eliminate restrictions such as the installation location.
- the programs corresponding to the processes performed by the detection device 104 (1), the detection device 104 (2), and the learning model generation device 106 are executed by a general-purpose computer or an integrated circuit such as an FPGA (Field Programmable Gate Array). You may do so. Further, the program may be recorded on a storage medium and provided, or may be provided through a network.
- a general-purpose computer or an integrated circuit such as an FPGA (Field Programmable Gate Array). You may do so.
- the program may be recorded on a storage medium and provided, or may be provided through a network.
- the object detection system and the object detection method according to the present invention in a system for detecting an object in a communication area from the propagation path information of a radio signal, a plurality of detection devices having different performances are installed at a plurality of places.
- the judgment data including the detection result of the detection device that performs highly accurate object detection as teacher data, it is possible to easily generate a learning model of object detection using supervised machine learning without spending time or manpower. And the generated learning model can be used in other detectors.
- 100 Object detection system; 101 ... AP; 102 ... STA; 103 (1), 103 (2) ... Capture device; 104 (1), 104 (2) ... Detection device; 105 ... network; 106 ... learning model generator; 150 ... object
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Remote Sensing (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Environmental & Geological Engineering (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Geophysics (AREA)
- Electromagnetism (AREA)
- Geology (AREA)
- Medical Informatics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
Description
図2は、AP101とSTA102との間で通信される無線信号のシーケンス例を示す。
図3は、本実施形態に係る物体検知方法のシーケンス例を示す。なお、図3に示すシーケンスは、図1の各装置により実行される。
図4は、検知装置104がキャプチャ装置103から取得するキャプチャ情報の一例を示す。なお、圧縮されたCSIが格納されたフレームと当該フレームの受信時刻の情報などを含むキャプチャ情報は、検知装置104の内部メモリなどの記憶部に蓄積される。
図5は、STA102からAP101に送信される圧縮されたCSIの一例を示す。図5において、左の列から順に、送信アンテナ数(AP101のアンテナ数)×受信アンテナ数(STA102のアンテナ数)、圧縮されたCSIの数、圧縮されたCSIの一例が記載されている。なお、送信アンテナ数は2以上である。
Claims (6)
- 2以上の無線装置間の伝搬路情報に基づいて、前記無線装置間の物体を検知する物体検知システムにおいて、
前記無線装置間の伝搬路情報をキャプチャして、キャプチャされた伝搬路情報を第1の検知装置および第2の検知装置に送信するキャプチャ装置と、
教師あり機械学習を用いた物体検知のための学習モデルを生成する学習モデル生成装置と
を備え、
前記第1の検知装置は、前記学習モデル生成装置により生成された前記学習モデルを用いて前記無線装置間の物体検知を行い、
前記第2の検知装置は、前記第1の検知装置よりも高精度な物体検知を行い、
前記学習モデル生成装置は、前記第2の検知装置の物体の検知結果を含む判定データを教師データとして前記第1の検知装置で用いる前記学習モデルの生成および更新を行う
ことを特徴とする物体検知システム。 - 請求項1に記載の物体検知システムにおいて、
他の無線装置間の伝搬路情報をキャプチャする他のキャプチャ装置をさらに備え、
前記第2の検知装置は、前記他のキャプチャ装置でキャプチャされた前記他の無線装置間の伝搬路情報を用いて前記第1の検知装置と同一範囲内で物体検知を行う
ことを特徴とする物体検知システム。 - 請求項1または請求項2に記載の物体検知システムにおいて、
前記第2の検知装置は、キャプチャされた伝搬路情報の全ての情報ではなく、一部の情報のみを利用して物体検知を行う
ことを特徴とする物体検知システム。 - 2以上の無線装置間の伝搬路情報に基づいて、前記無線装置間の物体を検知する物体検知方法であって、
前記無線装置間の伝搬路情報をキャプチャして、キャプチャされた伝搬路情報を第1の検知装置および第2の検知装置に送信するキャプチャ装置と、
教師あり機械学習を用いた物体検知のための学習モデルを生成する学習モデル生成装置と
を備え、
前記第1の検知装置は、前記学習モデル生成装置により生成された前記学習モデルを用いて前記無線装置間の物体検知を行い、
前記第2の検知装置は、前記第1の検知装置よりも高精度な物体検知を行い、
前記学習モデル生成装置は、前記第2の検知装置の物体の検知結果を含む判定データを教師データとして前記第1の検知装置で用いる前記学習モデルの生成および更新を行う
ことを特徴とする物体検知方法。 - 請求項4に記載の物体検知方法において、
他の無線装置間の伝搬路情報をキャプチャする他のキャプチャ装置をさらに備え、
前記第2の検知装置は、前記他のキャプチャ装置でキャプチャされた前記他の無線装置間の伝搬路情報を用いて前記第1の検知装置と同一範囲内で物体検知を行う
ことを特徴とする物体検知方法。 - 請求項4または請求項5に記載の物体検知方法において、
前記第2の検知装置は、キャプチャされた伝搬路情報の全ての情報ではなく、一部の情報のみを利用して物体検知を行う
ことを特徴とする物体検知方法。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US18/266,515 US20240036227A1 (en) | 2020-12-17 | 2020-12-17 | Object detection system and object detection method |
PCT/JP2020/047246 WO2022130585A1 (ja) | 2020-12-17 | 2020-12-17 | 物体検知システムおよび物体検知方法 |
JP2022569441A JP7435828B2 (ja) | 2020-12-17 | 2020-12-17 | 物体検知システムおよび物体検知方法 |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2020/047246 WO2022130585A1 (ja) | 2020-12-17 | 2020-12-17 | 物体検知システムおよび物体検知方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022130585A1 true WO2022130585A1 (ja) | 2022-06-23 |
Family
ID=82057460
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2020/047246 WO2022130585A1 (ja) | 2020-12-17 | 2020-12-17 | 物体検知システムおよび物体検知方法 |
Country Status (3)
Country | Link |
---|---|
US (1) | US20240036227A1 (ja) |
JP (1) | JP7435828B2 (ja) |
WO (1) | WO2022130585A1 (ja) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2010243231A (ja) * | 2009-04-02 | 2010-10-28 | Mitsubishi Electric Corp | 侵入物識別装置 |
JP2019100817A (ja) * | 2017-11-30 | 2019-06-24 | 株式会社東芝 | 位置推定装置、位置推定方法及び端末装置 |
JP2019158481A (ja) * | 2018-03-09 | 2019-09-19 | エヌ・ティ・ティ・コミュニケーションズ株式会社 | 判定システム、検知システム、判定方法及び検知方法 |
-
2020
- 2020-12-17 WO PCT/JP2020/047246 patent/WO2022130585A1/ja active Application Filing
- 2020-12-17 JP JP2022569441A patent/JP7435828B2/ja active Active
- 2020-12-17 US US18/266,515 patent/US20240036227A1/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2010243231A (ja) * | 2009-04-02 | 2010-10-28 | Mitsubishi Electric Corp | 侵入物識別装置 |
JP2019100817A (ja) * | 2017-11-30 | 2019-06-24 | 株式会社東芝 | 位置推定装置、位置推定方法及び端末装置 |
JP2019158481A (ja) * | 2018-03-09 | 2019-09-19 | エヌ・ティ・ティ・コミュニケーションズ株式会社 | 判定システム、検知システム、判定方法及び検知方法 |
Also Published As
Publication number | Publication date |
---|---|
US20240036227A1 (en) | 2024-02-01 |
JPWO2022130585A1 (ja) | 2022-06-23 |
JP7435828B2 (ja) | 2024-02-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10575275B2 (en) | Systems and methods for adaptively selecting distance estimates for localization of nodes based on error metric information | |
Zou et al. | Freedetector: Device-free occupancy detection with commodity wifi | |
EP1575325B1 (en) | Estimating the location of inexpensive wireless terminals by using signal strength measurements | |
Zhang et al. | A novel CSI-based fingerprinting for localization with a single AP | |
EP3316633B1 (en) | Apparatuses for generating a beacon sensor based network fingerprint | |
US9351179B2 (en) | Using one or more past channel estimates for adjusting parameters, sensing object presence, or allowing association | |
Liu et al. | A research on CSI-based human motion detection in complex scenarios | |
US10605889B2 (en) | Systems and methods for precise radio frequency localization using time sweep time difference of arrival | |
Li et al. | TDOA-based passive localization of standard WiFi devices | |
CN110999139A (zh) | 使用非连续或不连续通道精确射频定位的系统和方法 | |
WO2022130585A1 (ja) | 物体検知システムおよび物体検知方法 | |
Paulino et al. | Design and experimental evaluation of a Bluetooth 5.1 antenna array for angle-of-arrival estimation | |
Li et al. | Drone profiling through wireless fingerprinting | |
Paulino et al. | Evaluating a novel Bluetooth 5.1 AoA approach for low-cost indoor vehicle tracking via simulation | |
WO2021039516A1 (ja) | 空間画像生成装置、物体検知装置および物体検知方法 | |
WO2022130584A1 (ja) | 物体検知システム、物体検知方法、およびデータ振り分け装置 | |
CN114598994A (zh) | 一种映射已有蓝牙基站的方法及终端 | |
JP7209296B2 (ja) | 無線物体検知装置および無線物体検知方法 | |
Zimmermann et al. | Dynamic live wireless communication monitoring for jamming and interference detection in industry 4.0 | |
CN102692617A (zh) | 一种位置移动检测设备及其检测方法 | |
Boccadoroa et al. | Indoor Localization With Bluetooth: a Framework for Modelling Errors in AoA and RSSI. | |
CN111491368A (zh) | 一种适用于aoa算法定位基站的校正方法及校正装置 | |
Guo et al. | Rss-based localization using a single robot in complex environments | |
Puspitaningayu et al. | Accuracy investigations of fingerprint-based indoor localization system using ieee 802.15. 4 in two-floor environment | |
JP7459734B2 (ja) | 空間画像生成装置及び物体検知装置 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20965972 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2022569441 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 18266515 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 20965972 Country of ref document: EP Kind code of ref document: A1 |