Real-time positioning system and method that uses sensor-to-sensor data collection

Abstract

A real-time location system (RTLS) comprises sensors (512-514, 522-528, 532-542) and an upstream device that communicates with the sensors. Sensor pairs consist of a transmitting sensor (512) and a receiving sensor (514, 522-528, 532-542). Each sensor (512-514, 522-528, 532-542) is sequentially assigned as the transmitting sensor (512), while all other sensors are designated as the receiving sensors (514, 522-528, 532-542) and collect signal strength data. The upstream device determines a labeled dataset based on the signal strength data for the sensor pairs and zone designations associated with the signal strength data. The RTLS machine learning model is trained based on the labeled dataset.

Description

FIELD OF INVENTION

[0001] The present application relates to the field of asset tracking systems and, in particular, to a real-time location system of an array of radio frequency (RF) sensors for asset tracking. BACKGROUND

[0002] Building management systems encompass a variety of devices that support the monitoring and control of building operations. Building management systems often include one or more environmental control subsystems, such as security, fire protection, lighting, and heating, ventilation, and air conditioning ("HVAC") systems. The systems may also provide non-environmental control features for management and building occupants, such as tracking of assets (including occupants) indoors.

[0003] A real-time location system (RTLS) infrastructure installed in a building or a single floor plan typically consists of sensors installed at ceiling height, desk height, or both in a multi-zone area with varying sizes, ceiling heights, furniture, or other diverse geological features. RTLS can also be installed in multi-story buildings. The RTLS system tracks Bluetooth tags attached to assets. When a tag emits beacon signals, sensors installed in the building receive the beacons, and the characteristics of the received signal at the sensors (also known as sensor activation patterns) are used to locate the tags within the building floor plan.

[0004] RTLS system operation consists of two phases: an offline phase and an online phase. During the offline phase, a machine learning (ML) model is trained to recognize the characteristics of the received signal strength indicator (RSSI) for each zone.

[0005] A zone is simply any bounded area within a floor plan, such as a room. Training data is collected by moving tags in each zone, while RSSI values ​​are recorded at the sensors and the data is labeled with the zone name. During the online phase, the trained model is used to infer an RSSI signal coming from a tag to output the most likely zone location.

[0006] The data collection process during the offline phase is generally performed manually by a single person. This can be costly in terms of man-hours and logistically difficult to organize at customer sites, as it could disrupt the customer's operations. It is also inconvenient and may require customer security clearance. Summary

[0007] According to at least one embodiment of the disclosure, a real-time location system (RTLS) approach for asset tracking is provided. The RTLS uses machine learning techniques to locate asset tags within a facility. The approach describes an automated data collection method that does not require an agent to train an RTLS ML model.

[0008] One aspect is a real-time location system consisting of sensor pairs of multiple sensors and an upstream device that communicates with the sensors. The sensor pairs include a transmitting sensor and a receiving sensor for each sensor pair. In turn, each sensor is assigned as a transmitting sensor, while all other sensors are designated as receiving sensors, and signal strength data is collected at each receiving sensor. At the end of this process, the system ultimately has signal strength data for each sensor pair in the sensor network. The upstream device determines a tagged dataset based on the signal strength data for the sensor pairs and zone labels associated with the signal strength data. The upstream device supplements the signal strength data from the sensor pairs to simulate tags that transmit data to sensors at different locations in the zone to create a tagged dataset.The RTLS-ML model is trained based on the labeled dataset.

[0009] Another aspect is a method for a real-time positioning system. Signal strength data is collected for sensor pairs of multiple sensors. Each sensor pair includes a transmitting sensor and a receiving sensor. In turn, each sensor is assigned as a transmitting sensor, and all other sensors are designated as receiving sensors and collect signal strength data for the beacons transmitted by each transmitting sensor. A labeled dataset is determined based on the signal strength data for the sensor pairs and the zone labels associated with the signal strength data. The RTLS-ML model is trained based on the labeled dataset.

[0010] The features and advantages described above, as well as other features, will become more readily apparent to those skilled in the art from the following detailed description and the accompanying drawings. While it would be desirable to provide one or more of these or other advantageous features, the teachings disclosed herein extend to the embodiments falling within the scope of the appended claims, regardless of whether they achieve one or more of the above-noted advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] For a more complete understanding of the present disclosure and its advantages, reference is now made to the following descriptions taken in conjunction with the accompanying drawings, wherein like numerals designate like objects. Fig. 1 is a top plan view of an RTLS in an example implementation that may be used to employ the techniques described herein. Fig. 2 is a block diagram of exemplary components of a sensor of Fig. 1. Fig. 3 is a block diagram of exemplary components of an upstream device of Fig. 1. Fig. Figure 4 is a flowchart of the RTLS training and production-inference pipeline used to apply the techniques of Fig. 1 is operable. Fig. Figure 5 is a floor plan diagram illustrating an example data collection operation of the training-inference pipeline of Fig. 4 represents. Fig. 6 is a flowchart showing a data collection process of the training-inference pipeline of Fig. 4 in an exemplary implementation operable to apply the techniques described herein. Fig. 7 is a representation of a labeled dataset intended for the training process of RTLS-ML. DETAILED DESCRIPTION

[0012] Various technologies relating to systems and methods that facilitate the training and operation of a machine learning model for a real-time location system (RTLS) will now be described with reference to the drawings, in which like reference numerals represent like elements throughout the disclosure. The drawings discussed below and the various embodiments used to describe the principles of the present disclosure in this specification are for illustrative purposes only and should not be construed in any way to limit the scope of the disclosure. Those skilled in the art will appreciate that the principles of the present disclosure may be implemented in any suitably arranged apparatus. It is understood that functions described as being performed by particular system elements may be performed by multiple elements.Similarly, for example, one element may be configured to perform functions described as being performed by multiple elements. The numerous innovative teachings of the present application are described with reference to exemplary, non-limiting embodiments.

[0013] According to at least one embodiment of the disclosure, a real-time location system (RTLS) approach to asset tracking is provided. The RTLS uses machine learning techniques to locate asset tags within a facility. Conventional RTLS systems require an agent to move within each zone using wireless tags, recording data received from sensors, which consumes valuable time and resources. In some cases, these conventional systems can be logistically difficult to organize due to data collection during customer site operating hours and / or security concerns. The data collection approach described here uses sensor-to-sensor communication instead of mobile tag-to-sensor communication to generate training data for the RTLS-ML model.The RSSI-to-distance relationship is used to complement the signal strength of beacons transmitted by a transmitting sensor at a fixed location to a receiving sensor. This complement is used to simulate the beacon transmitted by a tag at different positions within a given zone.

[0014] Sensor activation is measured based on the received signal strength of communication between sensors, e.g., RSSI values. A label identifies a zone associated with a tag location. Labeled data, which includes the signal strength data and the zone labels, is used to train an ML model.

[0015] With reference to Fig. 1, a real-time location system (RTLS) 100 is shown for identifying the location of one or more assets within a facility. In some embodiments, the RTLS 100 may include a sensor management system that provides information to a building management system or other management services, such as a facility's environmental control system 102, to improve the services provided to building owners, managers, technicians, and occupants. The RTLS 100 may assist the building management system in managing environmental conditions, such as the conditions experienced by an occupant or other asset associated with the beacon tag, based at least in part on the locations of beacon tags 104. The RTLS 100 includes infrastructure 110, 112, 114 for determining locations of one or more beacon tags 104 within a structure, such as an interior or defined space of a facility.The structure may include various compartmentalized or designated areas 106, 108 of the structure, such as floors, rooms, corridors, or defined open spaces associated with the structure. Each area 106, 108 of the structure may include infrastructure devices, such as multiple sensors 110.

[0016] In some embodiments, a sensor hub 112 may be co-located with some or all of the sensors in an area 106, 108 of the device. In other embodiments, the sensor hub may be remote from the sensors and / or the device. For example, the sensor hub 112 may be located in the cloud and communicate directly or indirectly with the sensors 110, at least in part, via the internet or another communications network. The sensor hub 112 communicates directly or indirectly with the sensors 110, and the connection between the sensor hub 112 and the sensors may include wired, wireless, or both connections. Some embodiments of the RTLS 100 may include wired or wireless gateways 114 positioned between the sensors 110 in the device. The gateways may serve as communication transponders between the sensor hub 112 and the sensors.Similar to the sensors 110 and the sensor hub 112, the gateways 114 may also process data to use the techniques described herein.

[0017] The infrastructure sensors 110 can be mounted at fixed locations throughout the area 106, 108. The sensors 110 can be evenly distributed throughout each area 106, 108 or selectively distributed in a non-uniform manner, which may be preferred for beacon tag detection. The sensors 110 are configured to detect beacons emitted by one or more sources associated with the locations of assets. For example, residents can wear the beacon tags 104 for detection by the sensors 110, among other things, to facilitate tracking of the devices. Each of the beacon tags 104 can be any type of portable device capable of wireless communication with the facility sensors 110.Examples of beacon tags 104 include, but are not limited to, tags, wearables, wireless communication devices, tablets, portable computing devices, and any other type of portable device that includes circuitry for transmitting a beacon.

[0018] As in Fig. 1, the RTLS 100 includes a plurality of sensors 110 that can communicate with each other via wireless communication, such as radio frequency (RF). Examples of wireless communication include Bluetooth (including BLE), ultra-wideband (UWB), Wi-Fi (including Wi-Fi Direct), cellular, satellite, mesh networks, PAN, WPAN, WAN, Zigbee, near-field communication, and other types of radio communication and variations thereof. For any given sensor 110 of the RTLS 100, a transmit sensor 116 may be positioned at a transmit sensor location in the structure, and one or more proximal receive sensors 118, 120, 122, 124, 126, 128, 130 may be positioned at the receive sensor locations in the structure proximate to the transmit sensor.In some embodiments, one or more distal receive sensors 132, 134 may be positioned at receive sensor locations in the structure farther proximally from the transmit sensor 116, with one or more proximal receive sensors 118, 120, 122, 124, 126, 128, 130 located at least partially between the transmit and distal receive sensors or otherwise closer to the transmit sensor relative to the distal receive sensor(s). Each sensor and its location are within receive range of wireless signals transmitted by one or more of the other sensors. Although a particular sensor is described as a transmit or first sensor and specific sensors as receive or second sensors, it should be understood that all of the sensors of the RTLS 100 may be transmit and / or receive sensors.

[0019] For the RTLS 100, the transmit sensor 116 may transmit one or a burst of wireless signals that may be detected or received by one or more sensors 110, such as the proximal receive sensors 118-130, the distal receive sensors 132-134, or both. The receive sensors, such as the proximal or distal receive sensors 118-134, may determine a measured wireless signal ("wireless signal") and its corresponding value based on the received wireless signals. The measured wireless signal may, for example, be a received signal strength indicator (RSSI) or an equivalent value when a signal is transmitted, or statistical characteristics based on multiple received signals from the transmit sensor. Additionally, an expected wireless signal or expected statistical characteristics may be determined based on the location of the transmit sensor 116 (i.e., transmit sensor location) and the location of the proximal and / or distal receive sensors 118-134 (i.e.,Receive sensor location) or otherwise known to the receive sensors 118-134, a device upstream of the receive sensor (i.e., upstream device), or both. For example, from the location of these sensors, a distance between the transmit and receive sensors 116-134 can be determined. The distance can be correlated with the expected wireless signal (RSSI, etc.) or expected statistical characteristics. The locations of a given sensor pair correspond to an expected wireless signal from these sensors. Examples of the upstream device include, but are not limited to, the sensor hub 112 or the gateway 114.

[0020] The statistical features based on signals detected or received by the receiving sensor 118-134 or otherwise measured by the RTLS 100 are considered detected statistical features, and the statistical features determined by the RTLS 100 are considered expected statistical features. The detected and expected statistical features differ in that the expected statistical features are determined independently of the wireless signals detected or received at the second sensor 118-134 from the first sensor 116. The methods for determining each expected statistic may be based on, but are not limited to, a known distance between the first and second locations, a floor plan associated with the first and second sensors, or from a predefined lookup table associated with the first and second sensors.

[0021] Statistical features attempt to characterize the distribution of received wireless signals. Received statistical features and expected statistical features include primary statistical features and derived statistical features. Primary statistical features include, but are not limited to, a count of received wireless signals, a sum of received wireless signals, a squared sum of received wireless signals, and a combination of received signals from the wireless signals. Derived statistical features include, but are not limited to, the mean, standard deviation, mode, median, minimum, and maximum values ​​of the received wireless signal. The derived statistical features may be derived or calculated from the primary statistical features. In some embodiments, the separation of primary and derived statistical features for the primary purpose of bandwidth reduction may be considered somewhat artificial.For example, primary statistical features can be computed in one or more sensors and transmitted to an upstream device, and derived statistical features can be computed based on the primary statistical features to describe the statistical distribution of the wireless signals. If bandwidth concerns are minimal, received wireless signals from multiple sensors can be transmitted to the upstream device(s), allowing all or a significant portion of the statistical analysis to be performed upstream.

[0022] Statistical features may be determined in a receiving sensor and / or an upstream device. In particular, received or other statistical features may be determined in a receiving or second sensor 118-134, a gateway 114, the sensor hub 112 in the cloud, or a combination of these entities. Furthermore, the received wireless signals or a signal based on the received wireless signals may be passed to the upstream device (e.g., the sensor hub 112 and / or gateway 114) to determine the primary and / or derived statistical feature(s). For example, primary statistical features may be determined in a receiving sensor, and derived statistical features may be determined in the upstream device.The partitioning and coordination of statistical feature operations can be used to minimize network traffic and reduce the probability of transmission errors.

[0023] The upstream device 112, 114 may be configured to determine whether a difference between the measured wireless signal ("wireless signal") and the expected wireless signal and / or the corresponding values ​​exceeds predetermined thresholds and take certain actions. The actions may include, but are not limited to, notifying a remote device, communicating signals between sensors, or performing follow-up actions to determine the difference between the signal values. In one implementation, the upstream device 112, 114 may notify a mobile device associated with a technician or a control center of a building management system to investigate the sensors associated with the abnormal signal value difference.In another implementation, the upstream device 112, 114 may initiate further communications between the sensors 116-134, beyond the previously transmitted wireless signals from the transmitting sensor to the receiving sensor. For example, the transmitting sensor 116 may transmit wireless signals to other sensors (whether a proximal or distal receiving sensor 118-134) associated with the signal value difference, or vice versa. In doing so, the system may determine whether a sensor is in an expected position relative to the other sensor, corresponds to a position indicated in a floor or ceiling plan, or is otherwise functioning as expected. A non-RF signal, such as a light- or audio-based signal, may be used for further communication to compensate for any deficiencies in the RF signal previously transmitted between the sensors.In yet another embodiment, the upstream device 112, 114 may identify an obstruction between the sensors 116-134, such as a structural object, a moving object, or a living object. Examples of a structural object include, but are not limited to, a wall, partition, or overhang; examples of moving objects include, but are not limited to, furniture, equipment, or personal belongings; and examples of living objects include, but are not limited to, people, bugs, and other animals. In yet another embodiment, the upstream devices 112, 114 may identify one or more sensors in the environment 106, 108 that are malfunctioning, out of calibration, or are otherwise abnormally transmitting or receiving signals. These sensors may need to be reset, recalibrated, serviced, or replaced to maintain the integrity of the system. The RTLS 100 may detect obstructions in the structure, such as:the structural objects, moving objects, and living objects described above. By determining that a difference between the measured wireless signal and the expected wireless signal and / or the corresponding values ​​exceeds a specified threshold, the upstream device 112, 114 can determine that an obstruction exists between the transmit sensor 116 and the receive sensor (such as the proximal receive sensor 118-130 or the distal receive sensor 132-134). For example, as shown in FIG. Fig. 1, a first obstacle 136 may exist between the transmitting sensor 116 and the receiving sensor 122. A second obstacle 138 may exist between the transmitting sensor 116 and the receiving sensor 130. The signal value difference may indicate the obstacle(s) 136, 138, and the upstream device 112, 114 may take the appropriate action.

[0024] The RTLS 100 may expect one or more obstacles in the structure, particularly if indicated by a floor or ceiling plan. By determining that a difference between the measured wireless signal and the expected wireless signal exceeds a specified threshold, the upstream device 112, 114 may determine that it is inconsistent that the obstacle exists between the transmit sensor 116 and the receive sensor (such as the proximal receive sensor 118-130 or the distal receive sensor 132-134) in a situation where the obstacle should exist. For example, as in Fig. 1, a first obstacle 136 may exist between the transmit sensor 116 and the receive sensor 122. A second obstacle 138 may exist between the transmit sensor 116 and the receive sensor 130, or both. The signal value difference may indicate an absence of the obstacle(s) 136, 138 (i.e., may not indicate the existence of the obstacle), which may be caused, for example, by incorrect positioning of sensors or incorrect mapping of a sensor to a floor or ceiling plan. In response, the upstream devices 112, 114 may take appropriate action.

[0025] A sensor could exhibit signal differences between the measured wireless signal and the expected wireless signal, as explained above, that exceed thresholds or are otherwise determined to be persistently abnormal if multiple sensors are transmitting while the sensor is receiving. This could indicate that the sensor's reception capability is abnormal compared to the specification. When one sensor is transmitting, other sensors may exhibit signal differences that exceed thresholds or are otherwise determined to be abnormal. This behavior could indicate that the transmission capability is abnormal. In these cases, the upstream devices 112, 114 could also take appropriate action.

[0026] Fig. Figure 2 illustrates exemplary device components 200 of a sensor, such as sensor 110 of RTLS 100 for tracking beacon tags. The device components 200 of the sensor include a communication bus 202 for directly or indirectly connecting other device components. The other device components include one or more communication components 204 that communicate with other devices via a wired or wireless network, one or more processors 206, and one or more memory components 208.

[0027] The communication component 204 of the sensor is configured to communicate with the communication component of another sensor and receive a beacon broadcast from a beacon tag or another sensor. The communication component 204 may use wireless technology for communication, such as radio frequency (RF), infrared, microwave, lightwave, and acoustic communication. RF communication includes, but is not limited to, Bluetooth (including BLE), ultra-wideband (UWB), Wi-Fi (including Wi-Fi Direct), Zigbee, cellular, satellite, mesh networks, PAN, WPAN, WAN, near-field communication, and other types of wireless communication and variations thereof. In some embodiments, the communication component 204 of the device components 200 may also use wired technology for communication, such as transmitting data over a physical line, e.g., an electrical or fiber-optic medium.Each sensor may include more than one communication component 204. For example, a sensor may include a communication component for receiving the beacon from a beacon tag, a communication component for RF communication with another sensor, and a communication component for non-RF communication with the other sensor.

[0028] The processor 206 can execute code and process data detected or received by other components of the device components 200, such as information received in the communication component 204 or stored in the memory component 208. The code associated with the RTLS 100 and stored by the memory component 208 can include, but is not limited to, operating systems, applications, modules, drivers, and the like. An operating system includes executable code that controls basic functions of the sensor, such as interactions between the various components of the device components 200, communication with external devices via the communication component 204, and storage and retrieval of code and data to and from the memory component 208.

[0029] Each application includes executable code that provides specific functions for the processor 206 and / or the remaining components of the sensor. Examples of applications executable by the processor 206 include, but are not limited to, a communications module 210 for performing communications with other sensors, receiving transmissions from proximal beacon tags, and transmitting collected and / or calculated data to one or more upstream devices, and an operations module 212 for performing general and specific operations within the sensor. In some embodiments, the operations module 212 may determine a difference between the measured wireless signal and the expected wireless signal (or their corresponding statistical features and expected statistical feature values).The operation module 212 of each receiving sensor may also determine a signal strength value, a signal strength sum value, and a signal strength square value of a received beacon.

[0030] Data stored in the storage component 208 is information that can be viewed or manipulated by an operating system or application to perform functions of the sensor. Examples of data associated with the RTLS 100 and stored by the storage component 208 may include, but are not limited to, data values ​​214, such as expected wireless signals and measured wireless signals, and data thresholds 216, such as the predetermined threshold for comparison with the difference between the measured and expected wireless signals. The statistical data values ​​218 would include, for example, but are not limited to, the number of RF signals received, the sum of the RF signals received, the sum of squared RF signals received, and possibly the individual signal received.

[0031] The device components 200 of each sensor 110 may include one or more input components 220 and one or more output components 222. The input components 220 and output components 222 of the device components 200 may include one or more visual, audible, mechanical, or other components. In some embodiments, the input and output components 220, 222 of each sensor 110 may include a user interface 224 for interacting with a device user. The user interface 224 may include hardware and software to provide the user with the desired user experience. Examples of the input components 220 include, but are not limited to, a light sensor 226 (e.g., a passive infrared sensor), an audio sensor 228 (such as a microphone), a motion sensor 230, an atmospheric sensor 232 (such as temperature, humidity, pressure, etc.), and a combination of sensor technologies.Examples of output components 220 include, but are not limited to, displays 234, visual indicators 236, audio speakers 238, mechanical actuators 240, and a combination of production technologies. It is understood that . Fig. 2 is provided for illustrative purposes to show examples of the device components 200 of a sensor 110. It is not intended as a complete diagram of the various components used by the device. Therefore, the sensor 110 may include various other components described in Fig. 2 may comprise a combination of two or more components or a division of a particular component into two or more separate components and still fall within the scope of the present invention.

[0032] Fig. 3 illustrates exemplary components 300 of an upstream device, such as the sensor hub 112 or the gateway 114 for supporting RTLS. The device components 300 of the upstream device 112, 114 include a communication bus 302 for directly or indirectly connecting other device components. The other device components include one or more communication components 304 that communicate with other units via a wired or wireless network, one or more processors 306, and one or more memory components 308. The communication component 304 is configured to communicate with a plurality of sensors 110, either directly or via one or more other upstream devices. The communication component 304 may use wireless technology, wired technology, or a combination of these technologies to communicate with other devices, similar to the sensors discussed above with respect to Fig. 2 described.

[0033] The processor 306 can execute code and process data detected or received from other components of the device components 300, such as information received in the communication component 304 or stored in the memory component 308. The code associated with the RTLS 100 and stored by the memory component 308 can include, but is not limited to, operating systems, applications, modules, drivers, and the like. An operating system contains executable code that controls basic functions of the upstream device 112, 114, such as interactions between the various components of the device components 300, communication with external devices via the communication component 304, and storage and retrieval of code and data to and from the memory component 308.

[0034] Each application includes executable code that provides specific functions for the processor 306 or other components of the upstream device 112, 114, or both. Examples of applications that may be executed by the processor 306 include, but are not limited to, a data augmentation module 310 and a training / inference module 312. The data augmentation module 310 augments the sensor-to-sensor data set to generate the labeled data set. The training / inference module 312 trains an RTLS model and / or the location of a beacon tag using the trained RTLS model. The ML model 314 results from the training process and generates RTLS outputs based on the labeled data set.

[0035] The data stored in the storage component 308 is information that can be referenced and manipulated by an operating system or application to perform functions of the upstream device 112, 114. Examples of data associated with the RTLS 100 and stored by the storage component 308 include, but are not limited to, the training / inference data 316. The training / inference data 316 includes one or more labeled data sets and / or one or more beacon tag locations determined by the training / inference module 312. The storage component also includes the ML model 314, which is the result of performing the training operation 312 on the labeled data set 316.

[0036] The device components 300 of each upstream device 112, 114 may include one or more input and output components, i.e., I / O interfaces 318. The I / O interfaces 318 of the device components 300 may include one or more visual, audible, mechanical, or other components. In some embodiments, the I / O interfaces 318 of each upstream device 112, 114 may include a user interface 320 for interacting with a device user. The user interface 320 may include a combination of hardware and software to provide the user with the desired user experience.For example, the user interface 320 may include one or more input components to allow the user to enter information, such as mechanical buttons or an audio input, and one or more output components that provide information to the user, such as a visual or audible indicator to indicate an operating state of the upstream device 112, 114.

[0037] It is to be understood that Fig. 3 is only intended to illustrate examples of the device components 300 of an upstream device 112, 114 and is not intended as a complete diagram of the various components that the device may use. Therefore, the upstream devices 112, 114 may include various other components not shown in Fig. 3 may comprise a combination of two or more components or a division of a particular component into two or more separate components and still fall within the scope of the present invention.

[0038] With reference to Fig. Figure 4 illustrates a flowchart of the training and production reasoning pipeline 400 of the RTLS 100. The RTLS 100 includes a machine learning (ML) component, such as an analytics processor 410, that is trained on incoming data 420, namely a training dataset, and, after training, operates on the incoming data 420, namely an operational dataset. As represented by the reasoning pipeline 400, the system includes the analytics processor 410, which determines a labeled dataset 430.

[0039] If the incoming data 420 is the training dataset, the labeled dataset 430 is compared to the ground truth 440 and used during training to train the analytics processor 410. For example, the training dataset of the incoming data 420 is passed to the analytics processor 410 to help the processor learn to identify a specific zone 450 corresponding to the incoming data. In some embodiments, the analytics processor 410 is a machine learning model for real-time location services. For example, the analytics processor 410 may be an LSTM NN (long short-term memory neural network), which is used in the fields of artificial intelligence and machine learning. Supervised learning is used for the analytics processor 410 because the ground truth 440 is provided to the analytics processor 410. The ground truth is associated with a label to be generated by the machine learning model.

[0040] If the incoming data 420 is the operational data set, the labeled data set 430 identifies the locations of the beacon tags during operation. For example, the analytics processor 410 identifies a specific zone 460 based on the operational data set of the incoming data 420. For the production portion of the inference pipeline 400, the analytics processor 410 has been trained and is ready to determine the zone 460 of a beacon tag based on sensor RSSI values. The trained model of the analytics processor 410 is used to make inferences on live data to determine the tag's location. The model learns from its errors and can ultimately accurately predict the label, i.e., zone 460.

[0041] In some embodiments, the analytics processor 410 and its components may be located in the cloud. Sensors may send data as a dataset to the cloud via a gateway that sends the data to the cloud. The dataset is used to train the machine learning model of the analytics processor 410 in the cloud. The trained model may be deployed to the cloud, and the model may listen to sensor data and respond with a zone based on the received sensor data. With reference to Fig. 5 shows a floor plan 500 illustrating an example data collection process of the training-inference pipeline. The floor plan 500 is defined, and a plurality of zones 510, 520, and 530 are identified within the floor plan. Sensors 512-514, 522-528, 532-542 are then installed and commissioned in zones 510, 520, and 530 based on requirements regarding how many sensors are needed or desired per zone. A labeled dataset is then generated to train the machine learning component of the analytics processor 410. Specifically, signal strength data is collected for sensor pairs of multiple sensors 512-514, 522-528, and 532-542, each sensor pair including a transmit and a receive sensor.Each of the plurality of sensors 512-514, 522-528, 532-542 is assigned in turn as the transmitting sensor, and the signal strength data for each sensor pair corresponds to beacons transmitted by each transmitting sensor and received by each receiving sensor.

[0042] Collecting signal strength data can be illustrated by the following example, in which a transmitting sensor 512 transmits a beacon to reachable sensors 514, 522, 524, 526, 532, 534, 536, 538. As mentioned above, each sensor can act as a transmitting sensor in turn. Each sensor receiving the transmitted beacon can measure a signal strength value and generate RSSI statistics (mean, sigma, receive rate) based on the signal strength value, as previously described. The signal strength values ​​and / or RSSI statistics are provided to one or more upstream devices. The statistics can be used to verify the integrity of the sensor network before the data is used to train the ML model.

[0043] The upstream device generates an augmented, labeled dataset to train the ML-RTLS model. Sensors are used to send beacons to other sensors, unlike traditional systems that train machine learning models using a tag to send beacons to the sensors. To use data from a sensor to accurately represent data coming from a tag, the upstream device uses an RSSI-to-distance relationship to augment the sensor-to-sensor beacon RSSI. As a result, the augmented signal strength values ​​simulate data from a tag at various locations in the zone. For example, signal strength values ​​can be augmented to simulate various tag locations in a first zone 510 (Zone_0), such as the example location 550. An RSSI-to-distance model is used to generate the RSSI at simulated tag locations.In some embodiments, the RSSI distance relationship may be represented by:. Distance=10((measured power−RSSI) / (10*N))

[0044] Since the first sensor 512 may not detect its own beacon, the first RSSI value in the first sensor can be simulated using a second RSSI value from a sensor 514 in the same zone that is closest to the first sensor 512. The RTLS 100 determines a labeled data set to train the RTLS model. Specifically, the analysis processor adds a zone label to generate the labeled data set.

[0045] With reference to Fig. Figure 6 is a flowchart illustrating an exemplary process or system operation 600 of the RTLS 100. An ML model of an RTLS 100 may be trained using signal strength values ​​based on signals sent between sensors, rather than between a tag and a sensor. The ML model may learn a latent zone fingerprint from mapping signal strength data to zone labels. The ML model encodes or learns a fingerprint for each zone. To accurately represent data from a tag, the signal strength-distance relationship is used to augment the sensor-to-sensor signal so that it appears to be coming from a tag at different positions in the zone.

[0046] The process 600 of an RTLS includes a collection subprocess 610 and an analysis subprocess 630 for generating training data for the training-inference pipeline. Specifically, the analysis subprocess 630 determines a labeled data set based on the signal strength data for the sensor pairs and zone labels associated with the signal strength data. For the collection subprocess 610, signal strength data for sensor pairs of multiple sensors is collected (612, 614). Each sensor pair includes a transmitting sensor and a receiving sensor, and each sensor is assigned in turn as a transmitting sensor. Each sensor sequentially serves as the transmitting sensor and transmits (612) a beacon. Thus, each sensor is the transmitting sensor once. The other non-transmitting sensors receive (614) from the transmitting sensor in response to the beacon transmission (612).The signal strength data for each sensor pair corresponds to the beacons transmitted by each transmitting sensor and received by each receiving sensor. In some embodiments, each receiving sensor (616) can determine the signal strength value, a signal strength summation value, and a squared signal strength value of the received beacon. In either case, the receiving sensors (618) report data to the upstream device, whether it is the received measurement data or the calculated data determined from the measurement data. The collection subprocess continues until each sensor has an opportunity to transmit one or more beacons and the other sensors have an opportunity to receive the beacon (620).

[0047] For example, for the collection subprocess 610, a sensor in the network transmits a number of RF beacons at a predefined power level. All other sensors in the network listen and record the count, sum, and squared sum of the received RF beacons. The subprocess 610 is repeated with the next sensor in the network until all sensors in the RTLS have a chance to transmit, receive, and report. The collection subprocess 610 can be repeated multiple times to increase the robustness of its results.

[0048] For each sensor pair, a dataset collects signal strength values ​​and, in some embodiments, can calculate an RSSI sum value, a squared RSSI sum value, and an RSSI count value based on the signal strength values. The RSSI sum value is the sum of all received RSSI values. The squared RSSI sum value is the sum of the squares of all received RSSI signals. The RSSI count value is the number of received RSSI signals.

[0049] For the analysis sub-process 630 of operation 600, the sub-process generates statistics and labels. The analysis sub-process 630 generates (632) the signal strength statistics by sensor pairs, i.e., one sensor communicating with another sensor. The statistics can be used for various purposes, such as verifying the integrity of the sensor network before the data is used to train the ML model. In some embodiments, the signal strength statistics are generated by calculating (634) a mean feature, a standard deviation feature, and a received rate feature for each pair of transmit and receive sensors based on the signal strength values ​​of the wireless communication data. The mean feature, the standard deviation feature, and the received rate feature for each sensor pair can be determined based on the RSSI sum value, the RSSI sum square value, and the RSSI count value.The receive rate is calculated based on the number of beacons transmitted between the corresponding pair of transmitting and receiving sensors. The receive rate can be calculated by dividing the RSSI count by MAX_NUM_TX, which is the constant number of beacons transmitted by each beacon-emitting sensor.

[0050] In some embodiments, determining (632) the signal strength statistics includes filtering (638) the mean and standard deviation features based on the received rate feature. For example, the mean and standard deviation features may be filtered by assigning a zero value to a particular mean feature or standard deviation feature contingent on the corresponding received rate feature being below a threshold rate. The received rate data feature may be used to perform filtering (638) on the mean and standard deviation features. If the received rate for an entry is below a certain threshold, the data may be considered statistically insignificant and assigned the zero value.

[0051] In some embodiments, the analysis subprocess 630 of operation 600 determines operation of the upstream device to modify (640) the signal strength data to simulate a beacon tag position. The beacon tag position may be simulated based on signal strength statistics, a signal strength-versus-distance model, or both. In some embodiments, the upstream device (646) may simulate a signal strength value at a particular sensor based on a signal strength value of a nearest sensor that is closest to the particular sensor.

[0052] The analysis subprocess 630 of operation 600 determines (648) one or more labeled data sets, which may include sensor-to-sensor communication signal strength data and zone identifiers. The ML model of the RTLS 100 may be trained (650) using the labeled data set for the real-time location service. In some embodiments, the upstream device may determine a location of a beacon tag through the trained RTLS model of the real-time location system based on received signal strength data collected from a beacon tag in one or more sensors to determine a location of the beacon tag.

[0053] With reference to Fig. Figure 7 shows a labeled dataset 700 for the training process of the RTLS ML model. The labeled dataset 700 includes raw or processed signal strength data 710 collected by the receiving sensors. Zone labels 720 are added and assigned to each beacon signal received by the receiving sensors. The labeled dataset is used to train the machine learning component of the RTLS. Supervised learning is used to train the machine learning process of the analysis processor 510 to distinguish zones based on the RSSI fingerprint. Fig. In the embodiment shown in Figure 7, the labeled data is shown with the zone identified for each data point.

[0054] Those skilled in the art will appreciate that, for simplicity and clarity, the complete structure and operation of all data processing systems suitable for use with the present disclosure are not illustrated or described herein. Furthermore, none of the various features or methods described herein should be considered essential to any or all embodiments except as described herein. Various features may be omitted or duplicated in various embodiments. Various described processes may be omitted, repeated, performed sequentially, concurrently, or in a different order. Various features and operations described herein may be combined in still other embodiments, as may be described in the claims.

[0055] It is important to note that while the disclosure includes a description in the context of a fully functional system, it will be appreciated by those skilled in the art that at least portions of the mechanism of the present disclosure may be distributed in the form of instructions embodied in any machine-usable, computer-usable, or computer-readable medium. The present disclosure applies equally regardless of the type of instruction, signal-bearing medium, or storage medium used to carry out the implementation. Examples of machine-usable / readable or computer-usable / readable media include non-volatile, hard-coded media such as read-only memories (ROMs) or erasable, electrically programmable read-only memories (EEPROMs), and user-writable media such as floppy disks, hard disks, and read-only memories (CD-ROMs) or digital versatile disks (DVDs).

[0056] Although an exemplary embodiment of the present disclosure has been described in detail, it will be apparent to those skilled in the art that various changes, substitutions, variations, and improvements disclosed herein may be made without departing from the spirit and scope of the disclosure in its broadest form.

Claims

[1] Real-time positioning system using sensor-to-sensor data collection, comprising: Sensor pairs for multiple sensors, including a transmitting sensor and a receiving sensor for each sensor pair, wherein each sensor of the multiple sensors is sequentially assigned as the transmitting sensor, wherein all other sensors of the multiple sensors are designated as the receiving sensors and collect signal strength data for beacons transmitted by each transmitting sensor; an upstream device communicating with the plurality of sensors, the upstream device determining a labeled data set based on the signal strength data for the sensor pairs of the plurality of sensors and zone labels associated with the signal strength data; and an RTLS-ML model trained based on a labeled dataset. [2] The real-time location system of claim 1, wherein the labeled data set includes specific signal strength data for each sensor pair and a zone designation corresponding to the specific signal strength data for each sensor pair. [3] The real-time positioning system according to claim 1 or 2, wherein each receiving sensor of the plurality of receiving sensors determines the signal strength value, a signal strength sum value, and a signal strength square value of the received beacon. [4] A real-time location system according to any preceding claim, wherein the upstream device modifies the signal strength data to simulate a beacon tag position based on signal strength statistics and / or a signal strength-to-distance model. [5] A real-time location system according to any one of the preceding claims, wherein the upstream device simulates a signal strength value at a particular sensor of the plurality of sensors based on a signal strength value of a nearest sensor that is closest to the particular sensor. [6] A real-time location system according to any preceding claim, wherein the RTLS model is a machine learning neural network for identifying a zone associated with the signal strength data. [7] A real-time location system according to any preceding claim, wherein the upstream device determines a position of a beacon tag by the trained RTLS model based on signal strength data collected from a beacon tag and at least one sensor of the plurality of sensors to determine a location of the beacon tag. [8] A real-time location system according to any preceding claim, wherein the RTLS-ML model is trained to learn a latent zone fingerprint for each zone based on the labeled dataset. [9] A method for a real-time positioning system using sensor-to-sensor data collection, comprising: Collecting signal strength data for sensor pairs of a plurality of sensors, each sensor pair comprising a transmitting sensor and a receiving sensor, wherein each sensor of the plurality of sensors is sequentially assigned as the transmitting sensor, all other sensors of the plurality of sensors are designated as the receiving sensors, and collecting signal strength data for beacons transmitted by each transmitting sensor; Determining a labeled data set based on the signal strength data for the sensor pairs of the plurality of sensors, wherein the labeled data set includes zone labels associated with the signal strength data, and Train an RTLS-ML model to learn a latent zone fingerprint for each zone based on the labeled dataset. [10] The method of claim 9, wherein the labeled data set comprises specific signal strength data for each sensor pair and a zone designation corresponding to the specific signal strength data for each sensor pair. [11] The method of claim 9 or 10, further comprising: Determining the signal strength value, a signal strength sum value, and a signal strength square value of the received beacon. [12] The method of any one of claims 9 to 11, further comprising: modifying the signal strength data to simulate a beacon tag position based on signal strength statistics and / or a signal strength-to-distance model. [13] A method according to any one of claims 9-12, further comprising: Simulating a signal strength value at a specific sensor of the plurality of sensors based on a signal strength value of a nearest sensor that is closest to the specific sensor. [14] The method of any of claims 9-13, wherein the RTLS model is a machine learning neural network for identifying a zone associated with the signal strength data. [15] A method according to any one of claims 9-14, further comprising: Determining, by the trained RTLS model, a location of a beacon tag based on signal strength data collected from a beacon tag and at least one sensor of the plurality of sensors to determine a location of the beacon tag. [16] The method of any of claims 9-15, wherein training the RTLS model comprises training the RTLS-ML model to learn a latent zone fingerprint for each zone based on the labeled dataset.

Images