Monitor movable entities in a predefined area

Abstract

A method and system for enabling a computer to use real-time location information to determine reasons for occupancy of a predetermined area. The real-time location information is processed to determine occupancy information for each of a plurality of zones of the predetermined area. The (combined) occupancy information is processed using a machine learning model to predict reasons for occupancy of a predetermined zone of the predetermined area.

Description

Technical Field

[0001] The present invention relates to the field of monitoring movable entities and in particular to the field of determining the cause of occupancy of a predetermined area. Background Art

[0002] In the medical field, it is increasingly desirable to track and / or monitor the reasons why different zones of a predetermined area (such as a hospital or clinic) are occupied. This information is important for optimizing hospital workflows and identifying potentially inefficient operating practices.

[0003] One known method of identifying reasons for occupancy of an area is to have a dedicated observer monitor the area, for example in person or using CCTV, and identify events occurring within the area. The identified events can be considered as reasons for occupancy of different zones of the area. This method is overt and intrusive, as well as requiring additional expenses (e.g. substantial additional members of staff and / or equipment), and is therefore costly.

[0004] Another known method is to require the staff and / or other users of the area to record the reasons why they occupy one or more zones of the area. However, such a method is inaccurate, for example due to non-compliance or forgetfulness, and adds a significant burden to the users of the predetermined area.

[0005] Clearly, an automated and low-cost method of determining the cause of occupancy of a predetermined area would therefore be desirable. Summary of the invention

[0006] The invention is defined by the claims.

[0007] According to an example according to one aspect of the present invention, a computer-implemented method for monitoring a movable entity within a predetermined area is provided. The method includes: obtaining a real-time location RTL information stream identifying a current location of the movable entity within the predetermined area; processing the real-time location information stream to generate occupancy information for each of a plurality of predetermined zones of the predetermined area; and processing the occupancy information using a machine learning model to determine one or more occupancy reasons (29), each occupancy reason indicating an occurrence of a predetermined event within the one or more predetermined zones, wherein the machine learning model defines a relationship between the occupancy information and the occupancy reasons. .

[0008] The present invention recognizes that there is a relationship between the position of a movable entity and the occurrence of an event within the zone occupied by the movable entity. In other words, it has been recognized that the occurrence of a predetermined event in a predetermined area (formed by a zone) can be derived by monitoring the position of the movable entity with respect to the predetermined area.

[0009] The proposed embodiments provide a method for enabling mapping between the real-time location of a movable entity and a predetermined event associated with the movable entity. Advantageously, this provides a new input method for identifying an occupancy reason of a predetermined area, the occupancy reason being the occurrence of a predetermined event within the predetermined area.

[0010] In particular, the present invention recognises that occupancy information for zones of a predetermined area may be used to identify or predict the occurrence of a predetermined event for one or more zones (ie an event explaining why one or more particular zones are occupied by one or more movable entities).

[0011] In particular, the proposed method uses the real-time location of the movable entity to derive occupancy information for each of a plurality of predetermined zones. The occupancy information is then processed using a machine learning model to identify occupancy reasons for the plurality of zones.

[0012] In this way, the machine learning model is applied to the occupancy information of the plurality of zones to predict one or more occupancy reasons of the predetermined area. Thus, the machine learning model can map the occupancy information of the plurality of zones to the occupancy reasons. Advantageously, it has been recognized that the machine learning model can be trained to correlate the occupancy information of the plurality of zones with predetermined events occurring within one or more zones.

[0013] As used herein, an "occupancy reason" is data or information indicating the occurrence of an event within one or more zones (zones occupied by a movable entity), such as establishing an identification event that requires the presence of a specific movable entity or in particular results in the movement of a movable entity. Thus, an occupation reason may be the occurrence of an instance of a predetermined event that results in one or more specific zones being occupied by a movable entity. Thus, an occupation reason provides the reason why one or more zones are occupied, i.e. the purpose of occupying one or more zones.

[0014] In a particular embodiment, the occupancy reason may indicate the occurrence of an event that occurs within a particular or single zone. In other words, the step of determining one or more occupancy reasons may include identifying the occurrence of a predetermined event that occurs within a particular (single) zone. This provides more precise, specific and relevant information about the occurrence of the event.

[0015] The present invention provides a method for identifying the occurrence of a predetermined event (or multiple predetermined events) within one or more predetermined areas. This enables an (in)correct or (un)expected sequence of events to occur (e.g., by comparing the identified events with a schedule). The occurrence of incorrect events can indicate that the operating process is incorrect. In a clinical environment, this can be used to warn of the occurrence of incorrect medical procedures, which will significantly affect the health of the subject. In a manufacturing environment, this can be used to detect the occurrence of incorrect or unscheduled manufacturing steps, which will result in the manufacture of incorrect products.

[0016] In other words, detecting the occurrence of a predetermined event can be used, for example, to prevent an incorrect event from harming a patient in a clinical setting (e.g., an incorrect or unnecessary surgery) or an undesirable processing step in a manufacturing setting (e.g., an incorrect coating, cutting, or metal stamping process). Therefore, there is a technical motivation to provide and improve the identification of predetermined events within one or more zones.

[0017] Similarly, the proposed method can also ensure that desired events do occur, for example, to ensure that appropriate hygiene processes are performed prior to a surgical event (e.g., a cleaning event) or to ensure that the correct resources (e.g., personnel or equipment) are available prior to the next stage of the surgical process (i.e., the next event).

[0018] Here, movable entities include patients, clinicians, healthcare professionals, (medical) equipment (e.g., beds, defibrillators, ventilators, or anesthesia equipment), etc. Typically, movable entities are clinically relevant movable entities. Thus, movable entities may include movable (non-human) assets and / or persons / individuals. Thus, "movable entities" may alternatively be labeled "movable assets."

[0019] Therefore, some embodiments of the present invention provide a data processing architecture that allows a computer to be adapted to use real-time location information to identify occupancy reasons of a predetermined area. In other words, the proposed embodiments enable automatic monitoring of occupancy reasons of a predetermined area.

[0020] This enables determining occupancy reasons, such as the occurrence of "collaborative events" as described later, in a low-cost, sustainable, long-term, unobtrusive, accurate and private manner when compared to conventional means (such as interviews or observers).

[0021] Preferably, the occupancy information comprises non-binary information. In particular, the occupancy information may provide additional information in addition to merely indicating whether a zone is occupied. It has been recognized that real-time location (RTL) information may be used to create information with greater granularity than a simple binary indicator. Non-binary information may thus provide more contextual information, which helps improve the determination of the cause of occupancy. Some examples of suitable non-binary information for occupancy information will be provided hereinafter.

[0022] The computer-implemented method may be adjusted wherein: the occupancy information indicates a number of movable entities in each of the plurality of predetermined zones of the predetermined area; and the machine learning model defines a relationship between the number of movable entities within each predetermined zone and an occupancy reason.

[0023] Embodiments recognize that the number of movable entities within each zone may define the reason why one or more zones are occupied. Therefore, a machine learning model may be adapted to map the number of occupants in each zone to events occurring within one or more zones. Advantageously, this enables identification of predetermined events.

[0024] In some embodiments, the computer-implemented method includes determining one or more roles of each at least one movable entity, wherein: the occupancy information indicates the one or more roles of each movable entity in each of the multiple predetermined zones of the predetermined area; and the machine learning model defines a relationship between the one or more roles of each movable entity in each predetermined zone and the occupancy reason.

[0025] Preferably, there are at least two different possible types of roles or labels for the movable entity, more preferably at least three different possible types, and even more preferably at least five different possible types. The more possible types, the greater the granularity of the occupancy information, and therefore the greater the accuracy of the determined occupancy reason.

[0026] The role of a mobile entity (eg, the identity or capabilities of the mobile entity) may be used to help determine why a zone is occupied by the mobile entity. The proposed embodiments recognize that the roles or capabilities of mobile entities define the likely reasons why those entities occupy particular zones.

[0027] By way of example, occupancy information may be obtained from real-time location information provided by a real-time location system (RTLS). RTLS may be adapted to determine both the location of a movable entity and the role of the movable entity (eg, by tracking the location of movable entities having known roles).

[0028] Thus, embodiments provide a technique for evaluating occupancy reasons based on roles and zone locations of one or more movable entities in a predetermined area.

[0029] In a more preferred embodiment, the computer-implemented method includes determining one or more roles for each at least one movable entity, wherein: the occupancy information indicates the number of movable entities of each role in each of a plurality of predetermined zones of the predetermined area. Thus, the occupancy information may indicate the total number of movable entities of each role (or type) in each zone. Such instances of occupancy information (i.e., averaged at a single point in time or over a period of time) may be labeled as occupancy data entries. The occupancy data entry effectively comprises a histogram of roles and counts at a particular point in time or over a period of time.

[0030] Preferably, the step of processing the real-time location information stream includes generating a sequence of occupancy data entries as the occupancy information, each occupancy data entry indicating the number of movable entities within each of the multiple predetermined zones in the predetermined area at a single point in time; wherein each occupancy data entry is associated with a single point in time that is later than the previous occupancy data in the sequence of occupancy data entries, so that the entire sequence indicates the number of movable entities within each of the multiple predetermined zones within a time period.

[0031] The sequence of occupancy data entries thus indicates how the (zone) position of each movable entity changes over a period of time.Such an embodiment recognizes that occupancy reasons can be identified by monitoring how movable entities change position (e.g. at what speed or by evaluating the destination and / or origin).

[0032] The machine learning model may accordingly be adapted to identify or derive occupancy reasons by evaluating how the position(s) of the movable entity(ies) change over time (i.e. the sequence of movements).

[0033] Using a sequence of occupancy data entries (eg, snapshots of the position(s) of at least one movable entity) provides an intuitive and machineable mechanism for determining the cause of occupancy of (one or more zones of) a predetermined area.

[0034] Even more preferably, each occupancy data entry indicates the number of movable entities for each of a predetermined number of characters in each predetermined zone at a single point in time.

[0035] In other words, the method may include determining one or more roles of each at least one movable entity, wherein the step of generating the sequence of occupancy data entries includes generating a sequence of occupancy data entries, wherein each occupancy data entry indicates the number of movable entities of each role within each of the multiple predetermined zones of the predetermined area at a single point in time.

[0036] Thus, the occupancy data entries may indicate, for each of the predetermined zones, the number of movable entities of each of a plurality of different entity roles occupying the predetermined zone at a particular point in time.

[0037] Thus, the machine learning model may define a relationship between a sequence of occupancy data entries and one or more occupancy reasons (i.e., one or more predetermined events).

[0038] In some embodiments, the occupancy information comprises a plurality of sequences of occupancy data entries, and the step of processing the occupancy information comprises using a machine learning model to individually process each sequence of occupancy data to determine one or more occupancy reasons.

[0039] The use of a plurality of sequences enables an increased and / or more detailed determination of different occupancy reasons, ie different predetermined events.

[0040] Preferably, the plurality of sequences forms a series of sequences; and each of the sequences in the series is associated with a time period starting at a later point in time than an immediately preceding sequence in the series of sequences.

[0041] Thus, a sequence or series of sequences can be obtained, wherein each sequence is shifted in time from all other sequences in the series. This increases the time period over which the cause of the occupancy can be determined. Using multiple sequences also increases the precision with which the cause can be identified.

[0042] Preferably, each sequence is associated with a time period of predetermined length.

[0043] Optionally, each sequence in said series of sequences is associated with a time period overlapping with a time period associated with an immediately preceding sequence in said series of sequences.

[0044] By overlapping different sequences (in a series of sequences), it is less likely that an occupancy reason will be overlooked or not identified. This improves the accuracy of the method for determining an occupancy reason for a predetermined area.

[0045] By way of example, in case of no overlapping sequences - a predetermined event leading to occupancy of a specific zone spanning a time period overlapping two consecutive sequences may be ignored.

[0046] The method may include, after processing the real-time location information stream, discarding the real-time location information stream.

[0047] This improves the privacy of the movable entity since certain information about its whereabouts (i.e. real-time location) is discarded. This can also help improve the security of the area since such information cannot then be provided to another system or user - thereby reducing accidental leakage of information.

[0048] Preferably, determining one or more occupancy reasons comprises determining whether one or more predetermined events occurred during the time when the real-time location RTL information stream is provided.

[0049] In other words, the machine learning model may be adapted to map or associate the occupancy information with one or more predetermined events. That is, the machine learning model may be designed to identify the presence or impact of one or more predetermined events in the occupancy information.

[0050] The identification of scheduled events is important for understanding the reasons for occupancy of a scheduled area. In particular, it has been recognized that events (e.g., such as team meetings) can provide reasons why a zone of a scheduled area is occupied. By mapping occupancy information to one or more scheduled events, such reasons can be easily identified.

[0051] According to an example according to another aspect of the present invention, there is provided a computer program comprising code means for implementing any of the aforementioned methods when the program is run on a computer.

[0052] According to an example according to another aspect of the present invention, a system for monitoring movable entities in a predetermined area is provided. The system comprises: a stream receiving unit adapted to obtain a real-time location RTL information stream identifying a current location of a movable entity in the predetermined area; a processing unit adapted to process the real-time location information stream to generate occupancy information of each of a plurality of predetermined areas of the predetermined area; and an occupancy reason identification unit adapted to process the occupancy information using a machine learning model, thereby determining one or more occupancy reasons, each occupancy reason indicating the occurrence of a predetermined event in one or more predetermined areas, wherein the machine learning model defines a relationship between occupancy information and occupancy reasons.

[0053] The system may also include a role determination unit, which is suitable for determining one or more roles of each at least one movable entity, wherein: the occupancy information indicates the one or more roles of each movable entity in each of the multiple predetermined areas of the predetermined area; and the machine learning model defines the relationship between the one or more roles of each movable entity in each predetermined area and the occupancy reason.

[0054] Preferably, the processing unit is adapted to process the real-time position information stream to generate a sequence of occupancy data entries, each occupancy data entry indicating a number of movable entities within each of the plurality of predetermined zones of the predetermined area at a single point in time.

[0055] In some embodiments, the system is adapted wherein: the processing unit is adapted to generate multiple sequences of occupancy data entries; and the occupancy reason identification unit is adapted to use a machine learning model to individually process each sequence of occupancy data in the multiple sequences to determine one or more occupancy reasons.

[0056] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiment(s) described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] For a better understanding of the invention, and to show more clearly how the invention may be put into practice, reference will now be made, by way of example only, to the accompanying drawings, in which:

[0058] Figure 1 illustrates a predetermined area suitable for monitoring by a method according to an embodiment;

[0059] Figure 2 A method of monitoring a movable entity within a predetermined area according to an embodiment is illustrated;

[0060] Figure 3 illustrates a sequence of occupancy data entries for processing by a machine learning model to determine occupancy reasons;

[0061] Figure 4 It illustrates how to relate historical sequences to predetermined events;

[0062] Figure 5 illustrates the structure of a machine learning model according to an embodiment; and

[0063] Figure 6 A system for monitoring movable entities within a predetermined area according to an embodiment is illustrated. DETAILED DESCRIPTION

[0064] The present invention will be described with reference to the accompanying drawings.

[0065] It should be understood that the detailed description and specific examples, although indicating exemplary embodiments of the device, system and method, are intended only for illustrative purposes and are not intended to limit the scope of the invention. These and other features, aspects and advantages of the device, system and method of the present invention will become better understood based on the following description, the appended claims and the accompanying drawings. It should be understood that the drawings are only schematic and not drawn to scale. It should also be understood that the same reference numerals are used throughout the drawings to indicate the same or similar parts.

[0066] According to the concept of the present invention, a method and system are proposed, which enable a computer to use real-time location information to determine the occupancy reason of a predetermined area. The real-time location information is processed to determine the occupancy information of each of a plurality of zones of the predetermined area. The (combined) occupancy information is processed using a machine learning model to predict the occupancy reason of a predetermined zone of the predetermined area.

[0067] Embodiments are based at least in part on the recognition that real-time location information needs to be appropriately structured to be processed by machine learning models, and that such a structure is currently unavailable. Thus, by determining occupancy information for zones of a predetermined area, real-time location information can be given structure, thereby allowing detection of occupancy reasons.

[0068] For example, the illustrative embodiments may be employed in a hospital monitoring system to monitor and detect the occurrence of scheduled events within the hospital. The detection (or non-detection) of a scheduled event may be used to generate an alarm indicating whether a scheduled event did not occur, thereby reducing the occurrence of unscheduled / erroneous events and / or lost scheduled events.

[0069] The present invention is described in the context of a clinical environment. However, the embodiments may be extended to other environments, such as an industrial park, a warehouse, or a manufacturing plant.

[0070] Figure 1 Illustrated is a snapshot of a predetermined area 1 which has been conceptually divided into a plurality of predetermined zones 11-14.

[0071] The predetermined area 1 may be, for example, a specific floor or area of ​​a clinical building (e.g., a hospital or clinic), or may represent the entire clinical building. Each zone 11-14 of the predetermined area may represent a different room of the clinical building. For example, the first zone 11 may represent a ward 11, the second zone 12 may represent an examination room 12, the third zone 13 may represent a control room 13, and the fourth zone 14 may represent a corridor 14.

[0072] One or more movable entities 5, 6, 7 are distributed around a predetermined area. As illustrated, there may be more than one type of movable entity 5, 6, 7, wherein each type of movable entity is associated with a different role or identity. For example, a first movable entity 5 may be a nurse in a clinical environment, a second movable entity 6 may be a doctor in a clinical environment, and a third movable entity 7 may be a partner in a clinical environment. By way of further example, a fourth movable entity (not shown) may be a ventilation system for a patient, and a fifth movable entity (not shown) may be a defibrillator. Thus, each movable entity may identify a human entity (e.g., a specific clinician) or a non-human entity (e.g., a piece of medical equipment).

[0073] In other words, each movable entity may be associated with a particular role, which may be common to more than one movable entity (e.g., there may be multiple different nurses, each with the same "nurse" role). The role may define the capabilities, job role, training level, qualifications, experience, and / or other defining characteristics of the movable entity. For non-human movable entities, a "role" may be the use or purpose of the movable entity (e.g., a "bed" or "ventilator"), a severity classification (e.g., "high severity" for complex critical care equipment or "low" severity for vital signs monitoring equipment), etc. It will be clear that there are at least two different possible role types.

[0074] The position of each movable entity 5 , 6 , 7 is tracked, for example using a known real time location system (RTLS) 8 .

[0075] The real time location system 8 may be adapted to track when each individual movable entity 5, 6, 7 enters or leaves a predetermined zone 11-14 of the predetermined area 1. For example, the real time location system 8 may be formed by a network of tag / chain monitoring devices, each located at a passable boundary (e.g., a door) between different predetermined zones. An example of a suitable tag monitoring device is an access control tag reader (which allows or denies access to a movable entity via its tag). The real time location system may be adapted to give timing information indicating when a movable object has entered / left a zone.

[0076] Other methods of tracking the location of movable entities are known to the skilled person (e.g., monitoring where movable entities are checked in and out of an area). Other real-time location systems are also contemplated, such as those that can use wireless signals to track the location of an entity (even if the entity is not explicitly tagged (e.g., tracking a mobile phone)) or use closed-circuit television and image recognition software to track the location (and optionally the role) of an entity. Such systems can still report the location of an entity at regular intervals (the resolution can vary).

[0077] A real-time location system (RTLS) 8 generates a real-time location information stream that can be used to identify the current location of each movable entity 5, 6, 7 within the predetermined area 1. Specifically, the real-time location information may be able to identify when the movable entity leaves / enters the predetermined area (i.e., the time when the movable entity leaves / enters the predetermined area) and which predetermined area of ​​the plurality of predetermined areas the movable entity 5, 6, 7 is leaving / entering.

[0078] The real-time location information stream may also be able to indicate the type / role of each movable entity 5, 6, 7 that leaves / enters the predetermined zone. For example, control may be performed by issuing different types of movable entities with tags having different identities - or by assigning a tag with a unique identity to each movable entity and cross-referencing the role with each unique identity (e.g., using a database).

[0079] Table 1 illustrates an excerpt of a typical raw RTLS stream. The timestamp indicates the point in time at which the movable entity has had its location (e.g., exit and / or enter a zone) indicated, the tag ID identifies the movable entity 5, 6, 7, and the monitor ID identifies the tag monitoring device 8 that detects the movable entity. This can be used to detect the location of the entity. For example, if a monitor installed in room A (i.e., monitor ID = "room A") receives a signal from tag X (i.e., tag ID = "X"), it indicates that the entity associated with "tag ID = X" is in room A at the relevant timestamp.

[0080]

[0081] Table 1

[0082] Button Pressed indicates which button of multiple buttons (here: six buttons) was pressed when capturing the movement of the object. Typically, when one button is pressed, the corresponding button will be T. In the example where there are 6 buttons and button 5 is pressed, this field will be FFFFTF.

[0083] The tag type indicates the type of tag, such as whether it is a tag for tracking a patient (disposable wristband), tracking a staff member (badge-shaped), or tracking another entity (sticker-shaped). This can help identify the type of entity being monitored. In other words, different types of tags can be assigned to different types of entities to allow tracking of different entity types.

[0084] The real-time position information stream may thus indicate when the movable entity 5, 6, 7 has changed position or has moved to a different predetermined zone 11-14.

[0085] However, the typical real-time location information stream provided by RTLS is unstructured and cannot be reliably processed by machine learning models to identify occupancy reasons (which here refers to the occurrence of a predetermined event within a predetermined area).

[0086] The method described below provides a structure for real-time location information so that it can be processed by a (first) machine learning model to accurately determine the cause of occupancy.

[0087] Figure 2 A method 20 of monitoring movable entities 5, 6, 7 within a predetermined location 1 is illustrated.

[0088] The method 20 begins with step 21 of obtaining a real-time position information stream 28 identifying a current position of a movable entity within a predetermined area. Step 21 may include obtaining the real-time position information stream 28 from a real-time positioning system 8.

[0089] Subsequently, the method 20 performs a step 22 of processing the real-time position information to generate occupancy information for each of the plurality of predetermined zones 11 - 14 of the predetermined area 1 .

[0090] Step 22 may for example comprise generating an occupancy data entry comprising occupancy information for each of the plurality of predetermined zones.The occupancy data entry may indicate a current occupancy state of each of the predetermined zones, the current occupancy state being provided by the occupancy information.

[0091] In a first example, step 22 comprises identifying the number of movable entities 5, 6, 7 within each predetermined zone 11-14 at a particular point in time or the average number of movable entities 5, 6, 7 within each predetermined zone 11-14 over a period of time. Thus, the occupancy information for each predetermined zone may comprise a single value indicating the number of movable entities within the predetermined zone, such that the occupancy data entry comprises a corresponding value for each predetermined zone.

[0092] For example, this may include processing a real-time information stream to monitor the movement of each movable entity between different zones to identify the current location of each movable entity at a particular point in time. By way of example, movement of a movable entity from a first zone to a second zone indicates that the movable entity is located in the second zone. In this way, occupied zones may be identified, and the number of movable entities within an occupied zone may be identified. The number of movable entities in each predetermined zone may then be summed and output as occupancy information for each predetermined zone.

[0093] In a second example, step 22 includes determining what type / role each movable entity 5, 6, 7 is present in each predetermined zone 11-14 at a particular point in time. This may again include processing the real-time information stream to monitor the movement of each movable entity between the different zones, and to monitor which roles are present in each predetermined zone (e.g., via monitoring the identity of the movable entity). In this way, the role of the entity in each predetermined zone may be identified (e.g., via the identity of the movable entity).

[0094] Thus, there may be a step (not shown) of determining one or more roles for each at least one movable entity. For example, this may be performed by monitoring the identities (e.g., tag IDs) of the movable entities within a predetermined area and cross-referencing the identities to the appropriate roles. In another example, the role of the movable entity may be determined from the tag type (e.g., whether different tag types are assigned to movable entities of different types / roles). In some examples, both the identity and tag type of the movable entity may be processed (e.g., cross-referenced to a database) in order to establish the role / type of the entity.

[0095] In this way, the occupancy data entries may include information about the role of the movable entity in each predetermined zone.

[0096] In a third preferred example, step 22 comprises identifying in each predetermined zone 11-14 a plurality of movable entities 5, 6, 7 associated with each of a plurality of different roles / types.

[0097] Thus, as illustrated in Table 1 below, step 22 may comprise generating occupancy data entries indicating (for a particular instant or point in time) the number (Table 1: “Quantity”) of movable entities 5, 6, 7 of each of a predetermined number of types / roles (Table 1: “Entity Type”) in each predetermined zone 11-14 (Table 1: “Zone”).

[0098]

[0099] Table 1

[0100] In particular, Table 1 indicates the number of nurses 5, doctors 6, and companions 7 in each room 11-14 of the predetermined area 1. Thus, there is only one nurse 5 in the first predetermined area 11 (ward), two nurses 5 and one doctor 6 in the second predetermined area 12 (examination room), one companion 7 in the third predetermined area 13 (control room), and one doctor 6 in the fourth predetermined area 14 (corridor).

[0101] Therefore, the occupancy information obtained in step 22 preferably indicates at least a number of movable entities in each predetermined zone 11-14 at a specific point in time. Even more preferably, the occupancy information indicates the number of movable entities of each known role in each predetermined zone 11-14.

[0102] Thus, there may be a step (not shown) of determining one or more roles for each at least one movable entity. This information may be used to determine how many entities of each role are present in each predetermined zone.

[0103] Thus, the occupancy information may indicate, for a particular zone at a particular point in time, the number of movable entities of each type / role within the zone. However, other occupancy information may be obtained, for example, indicating the frequency (or number of times) at which movable entities of different types / roles moved in and out of a predetermined zone during a period of time immediately before the particular point in time.

[0104] Thus, step 22 includes processing the real-time location information stream so as to place the information contained therein in a structured format that can be processed by the machine learning model. In a specific example, a count of how many movable entities are in each predetermined zone is determined, and preferably a count of how many movable entities of each type are in each predetermined zone.

[0105] In particular, step 22 may comprise generating at least one occupancy data entry which provides a snapshot of occupancy information for each predetermined zone (ie at a particular point in time).

[0106] After the real-time location (RTL) information stream has been processed in step 22, the processed portion of the stream may be discarded (ie, no longer stored or deleted). This helps improve the privacy of the movable entity since the occupancy information may be anonymized (unlike the RTL information stream).

[0107] After step 22 , the method 20 performs a step 23 of processing the occupancy information using a (first) machine learning model to identify at least one occupancy reason 29 for the plurality of predetermined zones.

[0108] The (first) machine learning model is adapted to map occupancy information to occupancy reasons. In particular, the machine learning model may define a relationship between occupancy information and (at least one) occupancy reason.

[0109] The (first) machine learning model may be adapted to process the occupancy information for each zone as a whole (i.e., rather than processing each occupancy information individually) in order to determine the cause of occupancy.

[0110] As an illustrative example, the presence of all known movable entities within a single predetermined zone - such that no other zones are occupied - may indicate a team meeting (i.e., the occupancy reason is that a team meeting has occurred). In another example, the presence of all physicians and companions in a ward may indicate a traumatic event in the ward (e.g., a patient experiencing a cardiac arrest), which may be the only reason that all of the movable entities have a reason to occupy the ward - i.e., the occupancy reason.

[0111] In particular, step 23 of method 1 may include processing the occupancy information of each predetermined zone using a (first) machine learning model to identify whether a predetermined event has affected the occupancy information or the probability that a predetermined event affects the occupancy information. In other words, step 23 may include determining whether a predetermined event has occurred based on the occupancy information of each zone of the predetermined area.

[0112] In this way, the (first) machine learning model may be adapted to identify whether a specific predetermined event has occurred in the predetermined area 1, or to predict the probability that a specific predetermined event has occurred.

[0113] Thus, the machine learning model can be adapted to map or correlate occupancy information to events (e.g., surgeries or team meetings). In this way, events occurring in predetermined areas can be identified.

[0114] Step 23 may include processing the occupancy information using a plurality of different machine learning models, each machine learning model being adapted to determine whether a particular predetermined event has occurred in a predetermined area or the probability that a predetermined event has occurred.

[0115] Thus, step 23 may comprise using a corresponding plurality of machine learning models to check whether any of a plurality of predetermined events has affected or is associated with an occupancy data entry (including occupancy information for each predetermined zone).

[0116] Although it is possible to derive occupancy reasons using a machine learning model using data entries representing a single point in time / moment (as described above), in order to improve the predictive power of the machine learning model, a dynamic view of the occupancy information (i.e. over a period of time) can be used.

[0117] Thus, in a preferred embodiment, step 22 comprises obtaining a plurality of sequential occupancy data entries (i.e. a sequence of occupancy data entries), wherein each data entry is associated with a different point in time / instant. Data entries later in the sequence are associated with later points in time than entries earlier in the sequence. It will be apparent that each sequence of occupancy data entries is therefore associated with a defined time period - the time period spanned from the point in time associated with the first data entry to the point in time associated with the last / final data entry in the sequence.

[0118] Thus, step 23 may comprise processing the sequence of occupancy data entries using a machine learning model to determine or predict reasons for occupancy of one or more predetermined zones of the predetermined area.

[0119] In particular, step 23 may include using a machine learning model to determine whether a predetermined event has affected a sequence of occupancy data entries. For example, step 23 may include using a machine learning model to determine whether one or more predetermined events have occurred within a time window associated with or covered by the sequence.

[0120] Of course, more than one machine learning model may be applied, each machine learning model being adapted to identify whether a different predetermined event has occurred.

[0121] Examples of scheduled events (wherein the scheduled area is in a clinical environment) include: performance of a first or second type of surgery; team meeting; patient examination process; patient examination process; patient registration; consultation process; transport process; patient preparation process (e.g., in a waiting area); operating room preparation process (e.g., a disinfection process), etc.

[0122] It has been realised that a change in the position(s) of the at least one movable entity may be indicative of an occupancy reason(s) of the predetermined zone. This improves the accuracy of the step of determining the occupancy reason.

[0123] By way of example, in the case where a physician is moving from a pre-operative cleaning facility to a surgical procedure, such a situation may indicate that surgery will be performed (ie, surgery is the reason for the occupancy of the surgical procedure). Other examples will be apparent to those skilled in the art.

[0124] Each data entry in the sequence of occupancy data entries represents a different time sample of occupancy information of each of the plurality of predetermined areas. Therefore, each occupancy data entry in the sequence represents a different time point and reflects the overall occupancy state of the predetermined area at that time point.

[0125] Table 2 illustrates an example of a sequence of occupancy data entries formed by six different occupancy data entries (marked by numbers in the sequence - sequence numbers). Each data entry indicates how many mobile entities of each of a plurality of different types (entity types) of mobile entities 5, 6, 7 are in each predetermined zone 11-14 (zone) at a given moment. In Table 2 it can be clearly seen how the position of the mobile entities can change over time, i.e. as the mobile entities move around the predetermined area and enter into different zones.

[0126]

[0127] Table 2

[0128] The sequence generated in step 22 may be processed by a machine learning model in step 23 to identify reasons for occupancy of at least one zone of the predetermined area.

[0129] In particular, step 23 may comprise processing the sequence of occupancy data entries using one or more machine learning models to identify whether one or more predetermined events have occurred (in a predetermined area).

[0130] For example, each of multiple machine learning models can be tuned to determine whether a particular predetermined event has occurred or the probability that a particular predetermined event has occurred, where each machine learning model is adapted to identify a different event.

[0131] Even more preferably, step 22 may comprise generating a plurality of sequences of occupancy data entries. Each of the plurality of sequences may be offset in time from one another - i.e. forming a sequence / series of sequences. Each sequence in a series of sequences may be associated with a time period later than a time period associated with an earlier sequence in the series of sequences.

[0132] In other words, the step of generating a sequence of occupancy data entries may be performed iteratively to generate a plurality of sequences of occupancy data entries.Preferably, the plurality of sequences form a series of sequences arranged in the order in which they are generated.

[0133] Figure 3 A series 30 of sequences 31 - 34 that may be generated in step 22 of method 20 is illustrated.

[0134] Each sequence 31-34 has a predetermined length t pdEach consecutive sequence 31-34 is associated with a time period or time window that starts at a later point in time than an earlier sequence in the series.

[0135] In other words, the first sequence 31 may cover the first sequence 31 starting at a specific start time t st Start with a predetermined length t pd The second sequence 32 in the series covers a period of time starting at a specific start time t st The time interval t w The starting time period (predetermined length t pd )(i.e., t st +t w ). Each subsequent sequence follows the previous sequence by an additional time interval t w start.

[0136] Each sequence in the series itself comprises a sequence of occupancy data entries. As previously explained, each occupancy data entry may indicate occupancy information for each of a plurality of predetermined areas at a particular point in time or at a particular moment. Thus, each occupancy data entry may represent a sample of occupancy information for a predetermined area.

[0137] To form a series 30 of sequences, a sampling process may be performed. At a specified start time t st , can be per t i A predetermined number s of occupancy information samples are obtained every second to form a sequence (including s occupancy data entries and a sequence with a length of st i Then, at the specified start time t st The time interval t w At the beginning, the process can start again. The process can be repeated iteratively to obtain a series of sequences.

[0138] Thus, sequences may overlap with each other such that multiple sequences share one or more (but preferably not all) data occupation entries.

[0139] Each of the sequences of occupancy data entries in the series may be individually processed by the machine learning model to identify occupancy reasons during each time period covered by the corresponding sequence of occupancy data entries.

[0140] Once a sequence of occupancy data entries is available (i.e., all occupancy data entries for the sequence have been captured), processing by the machine learning model can be performed. This allows real-time or continuous detection of occupancy reasons, such as detecting ongoing scheduled events.

[0141] By overlapping sequences such as Figure 3), an improvement in continuous real-time detection of occupancy reasons may be achieved. By way of example only, a sequence may represent a time period lasting tens of minutes. Thus, if there is no overlap, a prediction can only be given every tens of minutes. However, allowing overlap (e.g., starting a new sequence every minute), then predictions can be made more frequently (e.g., every minute). This would allow us to give more frequent predictions to support real-time use cases.

[0142] It has been previously described how (one or more) machine learning models can process a sequence of occupancy data entries in order to identify whether a predetermined event has occurred, wherein the predetermined event forms an occupancy reason. In particular, the machine learning model can be adapted to identify the influence of the predetermined event on the sequence of occupancy data entries.

[0143] Machine learning models are well known in the art and typically operate by applying a series of algorithmic steps, layers, or masks to input data to predict output data. Typically, the output data indicates an answer to a predetermined question that the machine learning model is intended to answer (e.g., "Has a particular predetermined event occurred?"). Examples of machine learning models include neural network models and deep learning models.

[0144] A machine learning model is therefore any structure or process that enables the prediction of (the value of) certain output data based on input data - for example, calculating the probability that a predetermined event has occurred based on a sequence of input occupancy data entries.

[0145] Typically, to train a machine learning model, training data is collected that relates known examples of input data (e.g., sequences of occupied data entries) to known examples of output data (e.g., whether a particular predetermined event has occurred). The (generic) machine learning model is then applied to each example of the input data in an attempt to predict a corresponding example of the output data. The predicted examples of the output data are then compared to the known examples of the output data, for example by determining an error measure between the predicted examples and the known examples. Based on this comparison, the machine learning model is improved, with the purpose of improving the performance of the machine learning model (i.e., improving the accuracy of the predictions). This process can be repeated iteratively, thereby training or tuning the model to perform a specific task (e.g., identifying the occurrence of a predetermined event).

[0146] Methods of training, improving, or otherwise "optimizing" machine learning models, such as based on error metrics between predicted examples and known examples of output data, will be well known to those skilled in the art.

[0147] An embodiment of generating training data is described below, where the training data can be used to generate a machine learning model that processes a sequence of occupancy data entries to predict whether a single predetermined event has occurred or the probability that a single predetermined event has occurred. A skilled person will be able to adapt the methods described below to other examples of occupancy information.

[0148] However, training data that identifies known examples of sequences (and whether they are associated with known predetermined events) may not be readily available. Therefore, it may be necessary to generate appropriate training data formed by appropriate examples of sequences and their association or non-association with predetermined events.

[0149] Thus, an entry for the training data may indicate whether a given sequence corresponds to a given event.

[0150] In particular, each entry of the generated training data consists of a history sequence S (i.e., an example of a sequence that has occurred in the past) and an event label L e , the event label L e Indicates whether the historical sequence S and the scheduled event E correspond to each other (i.e., whether the historical sequence S indicates the scheduled event E). By way of example, the event label L e '0' may indicate that the scheduled event E did not occur when the historical sequence S occurred (ie, they do not correspond to each other), and the event label L e '1' may indicate that when the historical sequence S occurs (ie, they correspond to each other), the predetermined event E is ongoing. Of course, these labels may be reversed.

[0151] The skilled person will appreciate that different chunks / sets of training data may be generated for different predetermined events. Each chunk / set of training data may be used individually to train a different machine learning model to identify or predict the probability that a particular predetermined event (associated with the chunk / set of training data) is associated with a sequence of occupancy data entries.

[0152] In order to generate training data, multiple historical sequences and instances of predetermined events should be obtained, wherein each historical sequence and occurrence of the predetermined event has a defined start time and end time. Such data can be obtained, for example, by recording sequences of occupancy data and manually monitoring (e.g., in person or using CCTV) the start and end of the event while recording the sequences of occupancy data.

[0153] If the historical sequence completely overlaps with the time at which an instance of the predetermined event occurs, an entry in the training data may be generated indicating that the two are associated (e.g., with event label L eSimilarly, if there is no overlap in the time at which any instance of the historical sequence and the predetermined event occur, an entry for the training data may be generated indicating that the two are not associated (e.g., with event label L e '0').

[0154] In some embodiments, if there is a partial overlap between the historical sequence and the instance of the predetermined event, no training data is generated (ie, the information is discarded). This is to avoid inaccurately associating the historical sequence with the predetermined event, thereby improving the reliability of the training data.

[0155] According to well-known machine learning model training principles, this training data can be used to train or improve the machine learning model.

[0156] To provide additional entries for training data, if the sequence of occupancy data is within at least a predetermined period of time t ev In some embodiments, if the historical sequence partially overlaps with the instance of the predetermined event within a predetermined time period, then it can be determined that the historical sequence is associated with the predetermined event. In some embodiments, if the historical sequence partially overlaps with the instance of the event (but does not exceed the predetermined time period), data entries will not be generated for the training data.

[0157] Therefore, the event label L can be calculated for each sequence S for a specific predetermined event using the following relationship (where E represents an instance of a predetermined event): e :

[0158]

[0159] Therefore, it is possible to determine whether there is an overlap between the instance of the predetermined event and the historical sequence and whether the length of the overlap exceeds the predetermined time period t ev to generate training data.

[0160] exist Figure 4 An illustrative example of an instance of a historical sequence and a predetermined event is provided in . In particular, Figure 4 Different scenarios of historical sequences of occupancy data 41a-41e and instances of scheduled events 42a-42e are illustrated, which have the potential to form training data. Each historical sequence and instance of a scheduled event has a set start time and end time. In other words, the length and start time of each historical sequence and each instance of a scheduled event are known.

[0161] By way of example only, each instance of a predetermined event may represent a different instance of a particular surgical procedure (ie, each instance may represent a different procedure).

[0162] When the first sequence 41a starts and ends at approximately the same time point as the first instance 42a of the predetermined event, the first sequence 41a can be associated with the predetermined event. Therefore, an entry for training data (for a predetermined event) can be generated, including the first historical sequence and the event label L e '1'.

[0163] Similarly, the second sequence 41b can be associated with the predetermined event - because the second sequence 41b completely overlaps with the second instance 42b of the predetermined event (i.e., as the second sequence occurs, the instance of the predetermined event is ongoing). Therefore, the second sequence and the predetermined event can be associated with each other. Therefore, an entry for training data (for the predetermined event) can be generated and includes the second historical sequence 41b and the event label L e '1'.

[0164] However, the third sequence 41c cannot be associated with the predetermined event because there is no overlap between the third sequence and the third instance 42c of the predetermined event. Therefore, an entry for training data (for a predetermined event) can be generated, including the third historical sequence 41c and the event label L e '0'.

[0165] The fourth sequence 41d and the fourth instance 42d of the predetermined event 42d partially overlap each other. The length of time t during which they partially overlap each other is o1 Greater than the predetermined time length t ev Therefore, the fourth sequence and the scheduled event can be associated with each other. Therefore, an entry for training data (for a scheduled event) can be generated, including the fourth historical sequence 41d and the event label L e '1'.

[0166] The fifth sequence 41e and the fifth instance 42e of the predetermined event also partially overlap with each other; however, the overlapping period t o2 Not longer than the predetermined time length t ev Therefore, it cannot be assumed that the fifth sequence 41e and the predetermined event correspond to each other. However, it cannot be confirmed that there is no correspondence between the fifth sequence 41e and the predetermined event (because the overlap is greater than 0). Therefore, in order to ensure accurate training data, in this case, no data entry for training data is generated.

[0167] In this way, training data is generated that associates historical sequences of occupancy data entries with predetermined events. This training data can be used to train a machine learning model.

[0168] In one example of a training process, N sets of training data can be generated using the above process (for different scheduled events) to train N machine learning models, each machine learning model being associated with a different scheduled event. Thus, N machine learning models can be trained individually for each of the N scheduled events.

[0169] In one example, each data entry of the training data used to train the machine learning model includes a historical sequence of occupancy data entries and an event label '1' or '0'. As described above, each historical sequence of occupancy data entries may include, for example, a matrix having f columns and l rows. Each row represents a different instance of the occupancy data entry (i.e., a different sample of occupancy information). The columns may be formed by X groups of Y columns (each group containing occupancy information, and each group is associated with a predetermined zone - i.e., X indicates the number of predetermined zones), wherein each column in a group indicates the number of movable entities of a specific role in the predetermined zone (i.e., Y indicates the number of different roles of the movable entity). Therefore, the number of columns in the matrix is ​​equal to the product of X and Y.

[0170] Event labels can thus be used as model training targets for training machine learning models.

[0171] Figure 5 The structure 50 of a machine learning model for predicting (the probability p of) the occurrence of a predetermined event based on a sequence 51 of occupancy data entries according to an embodiment of the present invention is illustrated.

[0172] The structure 50 includes a single input layer 52 to a sequence 51 in matrix form (e.g., as illustrated in Table 2). The size of the matrix is ​​fixed (i.e., the sequence of occupied data entries is in a fixed format). The input layer feeds into a set 53 of one or more hidden layers 53a-53c, each with an LSTM (Long Short Term Memory) and a dropout node. The output of the set of hidden layers is fed into an output layer 54 using a single dense neuron. The dense neuron can generate a floating point or decimal number between 0 and 1, which indicates the probability that a predetermined event is occurring during the time window of the sequence 51 input to the machine learning model.

[0173] In order to evaluate and improve a machine learning model for identifying (the probability of) predetermined events, an evaluation framework is preferably used.

[0174] The evaluation framework may include dividing the training data for specific predetermined events into different groups, and using different groups of training data for different purposes.

[0175] In certain embodiments, the training data may be divided into three different groups: a training group; a validation group, and a test group.

[0176] The training set of training data is used to train and adjust the machine learning model to predict whether a specific predetermined event (its probability) has occurred in the time window covered by the sequence of occupied data entries. The training of the machine learning model can use any standard training method, such as stochastic / incremental gradient descent (SGD) or the "Adam" algorithm. The validation set is used to tune the parameters of the machine learning model to ensure that the machine learning model is not just biased towards the test set. The test set is used to evaluate the accuracy or performance of the machine learning model to evaluate whether the machine learning model has been successfully adjusted so that it can accurately predict whether the predetermined event has occurred (its probability) within the time window covered by the sequence of occupied data entries.

[0177] The performance of the machine learning model can be evaluated using any known learning classification metric, such as accuracy, precision, recall, ROC (receiver operating characteristic) curve, and AUC (area under the curve). However, since it is recognized that the training data of the present invention is usually skewed (because there are usually more '0's than '1's in the labels), it is recommended to use a metric that is insensitive to skewness, such as AUC.

[0178] The ratio of splitting the training data into three different groups is flexible, such as 4:3:3 or 5:3:2, respectively. It is recognized that sequences covering time windows close to each other (of training data) will likely have similar event labels. Therefore, adjacent sequences should not be split across different groups (as this may lead to an erroneous assessment of the accuracy of the machine learning model).

[0179] For example, the training data may cover a period of 10 days, wherein the first 4 days may be used as a training set; days 5, 6, and 7 may be used as a validation set, and days 8, 9, and 10 may be used as a testing set.

[0180] The training method described above has been directed to an embodiment in which a machine learning model is trained to identify whether a single predetermined event occurs (or a probability) during a time window covered by a sequence of occupancy data entries - each occupancy data entry is for each of a plurality of zones of a predetermined area, including occupancy information for the zone. In particular, the trained machine learning model is adapted to process a sequence of occupancy data entries to identify whether a predetermined event occurs or a probability of occurrence during a time window associated with the sequence.

[0181] However, the training method can be adapted to other machine learning models, for example, to process other input data (such as individual occupancy data entries) to identify occupancy reasons, i.e., the occurrence of predetermined events.

[0182] Scheduled events may alternatively be labeled "cooperation events" because they typically require the presence of more than one movable entity (ie, the entities cooperate in order to perform the scheduled event).

[0183] Based on the continuous processing of occupancy information, the proposed method can be employed to monitor the ongoing occupancy reasons (eg, whether a predetermined event is currently occurring).

[0184] In at least one embodiment, this continuous monitoring can be used to generate an alarm when an incorrect occupancy reason (i.e., a scheduled event) is occurring. In particular, the detected occupancy reason can be compared with the scheduled occupancy reason, and if the two do not match, an alarm can be generated.

[0185] For example, if a first scheduled event (e.g., surgery on a patient occupying a first patient room) is scheduled to currently occur, but the machine learning model indicates that a second, different scheduled event (e.g., surgery on a patient occupying a second patient room) is currently occurring, an alert can be generated to indicate that an incorrect or unscheduled scheduled event is occurring.

[0186] Similarly, if a first scheduled event is scheduled, and it is determined that the first scheduled event is not ongoing, an alert may be generated to identify that the first scheduled event has not occurred.

[0187] In this way, the proposed embodiment provides automatic detection of whether a movable entity in a predetermined area complies with a scheduled task or event by processing the real-time location of the movable entity. If the movable entity does not comply with the scheduled task or event, one or more alarms may be generated.

[0188] Such proposed embodiments may help prevent improper events (eg, improper manipulation of a patient) from occurring and reduce errors in performing tasks in a predetermined area.

[0189] The alarm may be in the form of a visual, audio or tactile alarm, for example, or may be an electronic alarm (eg a log or flag in the data).

[0190] In another example, if a first scheduled event (e.g., a first medical examination) is completed earlier than expected (e.g., 10AM instead of 11AM), an alert can be generated to suggest that a subsequent scheduled event (e.g., a second medical examination) can start earlier, or that the relevant room is free for additional events (e.g., a third medical examination or a cleaning process). Such an example can result in higher room utilization (more examinations can be scheduled), thereby reducing patient waiting time.

[0191] In an embodiment, a display (e.g., screen) may be provided to display the automatically determined reasons for occupancy. This improves awareness of potential customers, staff, and clients without requiring the personnel to request (e.g., from potentially busy entities) information about the progress of an event.

[0192] For example, a display can be set up in the waiting area of ​​an operating room. The machine learning model can automatically determine the scheduled events occurring in the operating room (e.g., room being prepared, patient being transported, operation in progress, handling a package, etc.) and display the reason for occupancy on the screen so that the subject, future patients, and / or hospital staff can know the progress / status of the operation without having to enter the operating room or ask the physician / surgeon who has left the operating room.

[0193] Thus, embodiments may also include a display for displaying the reason for occupancy for viewing by the subject and / or others.

[0194] In another example, the determined occupancy reasons can be communicated to a reporting system of the hospital (or other predetermined area). In this way, management personnel can have automatic reports of room usage statistics, such as the distribution of times when different examination times occur, without having to hire personnel to observe this data because it can be collected automatically. Thus, the efficiency of use of the predetermined area can be increased.

[0195] In addition to identifying the occurrence of a predetermined event within one or more predetermined zones (ie identifying an occupancy reason), some methods may include the further step of estimating / determining the remaining time or progress of the (ongoing) predetermined event.

[0196] In a simple embodiment, this can be performed by the following steps: identifying the start time of an ongoing scheduled event, which can identify the time of the ongoing scheduled event based on the machine learning method previously described; determining the amount of time that has passed since the start time (e.g., the difference between the current time and the start time); and determining the difference between the amount of time that has passed and the predicted length of time for executing the ongoing scheduled event.

[0197] The predicted time length for executing the ongoing scheduled event can be calculated from a predetermined value for the length of the scheduled event (e.g., listed in a lookup table). By way of example, each possible scheduled event can be associated with a corresponding predicted time length, e.g., based on historical averages / trends or clinical guidelines / recommendations.

[0198] In more complex embodiments, estimating / determining the remaining time of a (ongoing) scheduled event may include determining the current sub-phase of the scheduled event by processing the occupancy information. In particular, each scheduled event may be divided into a plurality of different sub-phases, which may be identified using machine learning methods. Once the current sub-phase of the ongoing scheduled event is identified, the remaining length of time to complete the scheduled event (e.g., to complete the remaining sub-phases) can be predicted, which may again be based on historical averages / trends and / or clinical guidelines / recommendations.

[0199] In some embodiments, a second machine learning method is used to estimate / determine the remaining time of an ongoing scheduled event. This can be performed in a conceptually similar manner to the aforementioned machine learning method.

[0200] For each scheduled event, there may be a corresponding second machine learning method, and an appropriate second machine learning method may be selected to estimate the remaining time of the corresponding ongoing scheduled event.

[0201] In one example, a second machine learning method may receive as input an identification of a scheduled event detected by an earlier machine learning method and an elapsed time of the scheduled event, and output a prediction of the remaining time to the detected scheduled event.

[0202] In an example, a first machine learning method may process occupancy information for each of a plurality of predetermined zones to detect an occurrence of a predetermined event, and a second machine learning method may process the same occupancy information to determine a remaining time for the detected predetermined event.

[0203] The occupancy information may comprise a sequence of occupancy data entries, each occupancy data entry providing occupancy information for each of a plurality of predetermined zones (as previously described).

[0204] In some embodiments, the second machine learning method may use the elapsed time of the scheduled event as an additional input, ie, together with the occupancy information, for determining the remaining time of the scheduled event.

[0205] One example uses an extended version of the previously described occupancy data entries to provide data suitable for predicting the remaining time of an ongoing scheduled event. Specifically, a method includes: identifying the start time of the ongoing scheduled event (e.g., identifying the time of the ongoing scheduled event); determining the time elapsed from the start time of each occupancy data entry, and providing each occupancy data entry with the time elapsed since the start time of the associated scheduled event. The elapsed time can be calculated, for example, by determining a start timestamp for the start of the current event and calculating the elapsed time using the start timestamp and the timestamp of the occupancy data entry. A machine learning method can then be used to process one / more occupancy data entries to predict how much time is left for the scheduled event.

[0206] It will be appreciated that each occupancy data entry may be associated with a particular identified event (ie an event determined to have occurred during that occupancy data entry).

[0207] It has been previously described how to detect a scheduled event by processing a sequence of occupancy data entries using a machine learning method. This sequence of occupancy data entries can be modified to provide a suitable input for a machine learning method for predicting the remaining time for completing a scheduled event.

[0208] In some embodiments, a sequence of occupancy data entries may be formed, each occupancy data entry in the sequence further comprising a data field indicating the length of time that has elapsed since the event (corresponding to the occupancy data field) began (or was first detected). This forms a "sequence of modified occupancy data entries".

[0209] The step of determining the time remaining for the ongoing scheduled event may include processing the sequence of modified occupancy data entries using a second machine learning method to determine how much time remains for the scheduled event.

[0210] Table 3 illustrates an example of a modified sequence of occupancy data entries, which is the same as Table 2 (for predicting the occurrence of a predetermined event - i.e. a predetermined event associated with a sequence of occupancy data entries), with an additional data field for each occupancy data entry indicating the time that has elapsed since the start of the predetermined event (or since it was detected that the predetermined event is occurring).

[0211]

[0212]

[0213] Table 3

[0214] The inventors have recognised that the progress of a scheduled event affects the movement of a movable entity within a predetermined zone. Therefore, the remaining time of an ongoing scheduled event can be more accurately predicted by monitoring the movement of the movable entity (eg using a sequence of occupancy data entries).

[0215] By way of example only, it can be appreciated that for a certain portion of a scheduled surgical procedure, two surgeons (rather than one) are required. Thus, the entry of a second surgeon into a particular zone can indicate that a particular portion of the scheduled surgical procedure has begun, thereby indicating the current progress of the surgical procedure (which affects the remaining time of the surgical procedure).

[0216] The calculated remaining time may be presented, such as via a user interface.

[0217] A brief description of the machine learning approach has been provided previously. The second machine learning approach employs a similar principle.

[0218] An embodiment of generating training data for a second machine learning method is described below, wherein the training data can be used to generate a second machine learning model that processes a sequence of modified occupancy data entries to predict the remaining time of a scheduled event. Those skilled in the art will be able to adapt the method described below to a machine learning method suitable for predicting the progress of a scheduled event.

[0219] The expected training data for the second machine learning model will provide as input examples a sequence of modified occupancy data entries (each data entry providing relevant occupancy information and elapsed time), and provide as output examples the remaining time to a scheduled event (associated with the modified sequence).

[0220] Typically, available information for preparing suitable training data will include example sequences of occupancy data entries and known occurrences of scheduled events (along with start and end times for both). Typically, scheduled events are manually labeled.

[0221] As previously mentioned, this available information can be obtained by obtaining a plurality of historical sequences, and instances of predetermined events should be obtained, wherein each historical sequence and occurrence of a predetermined event has a defined start time and end time. Such data can be obtained, for example, by recording a sequence of occupancy data and manually monitoring (e.g., in person or using CCTV) the start and end of an event while recording the sequence of occupancy data. As previously described, Figure 4 This is illustrated in .

[0222] To generate training data for the second machine learning method, appropriate instances should be selected for the input examples of the training data. In other words, only some of the sequences of occupied data entries may provide suitable training examples for calculating the remaining time to the predetermined event, for example, by being sufficiently associated with or indicating the occurrence of the predetermined event.

[0223] Initially, if a sequence of occupancy data entries does not overlap with a predetermined event, the sequence may be discarded. This would mean that only those sequences that are temporally related to the predetermined event are selected to form the training data.

[0224] In the second step, if a sequence of occupied data entries has an end time at or before the end time of the overlapping scheduled event, then the sequence can be selected as an input example. This provides an input example according to which there is some remaining time before the associated scheduled event ends.

[0225] In some embodiments, to provide additional input examples, a threshold t may be set, and sequences that meet certain criteria (e.g., the end time t of the sequence) may be selected. 结束(s) The end time t of the overlap with the scheduled event (completed later) 结束(e) The difference between is equal to or greater than t) can also be selected as an input example.

[0226] Having established suitable input examples for the training data, it is necessary to generate corresponding output examples for each input example. This can be performed by simply determining the difference between the end time of a predetermined event that overlaps with the sequence (forming the input example) and the end time of the sequence. In another embodiment, the following relationship may be used:

[0227]

[0228] This provides a suitable example output for each (selected) example input, thereby generating training data for the second machine learning method.

[0229] In some embodiments, multiple second machine learning methods are generated, one for each predetermined event type. A skilled person will be able to easily generate appropriate training data for each of the multiple predetermined events (e.g., by limiting the selection of sequences to those that overlap with the corresponding predetermined event).

[0230] Methods for training machine learning algorithms have been previously described, and can be adapted to train a second machine learning method (using appropriate training data).

[0231] Although the previous embodiments have been described in the context of predicting the remaining time of a scheduled event, the embodiments also extend to methods for predicting the progress (e.g., progress percentage) of a scheduled event. Therefore, references to the term "remaining time" may be replaced by the term "progress" by analogy.

[0232] Of course, it will be appreciated that it is not necessary to use the methods described herein for identifying the occurrence of a predetermined event. Rather, any method for identifying the occurrence of a predetermined event from occupancy information may be utilized and then processed to identify the remaining time of the detected predetermined event, for example, according to the described examples.

[0233] Although the above embodiments are described in the context of the movable entity being a healthcare professional, it will be appreciated that the embodiments may be extended to monitoring other persons - such as patients or visitors - as well as monitoring inanimate movable objects - such as beds, medical equipment, etc. Thus, the movable entity may be any animate or inanimate movable entity / asset within a predetermined area.

[0234] As briefly described previously, it will also be appreciated that the predetermined area need not be in a clinical setting. For example, the predetermined area may be an industrial site (e.g., a factory), wherein different zones represent different areas of the industrial site. Possible reasons for occupancy (predetermined events) of a factory may indicate, for example, a specific stage of a manufacturing process (e.g., a coating, cutting, or stamping stage). Other examples will be readily apparent to the skilled person.

[0235] Figure 6 A system 60 for monitoring movable entities within a predetermined area is illustrated.

[0236] The system 60 comprises a stream receiving unit 61 adapted to obtain a real-time location RTL information stream 65 identifying the current location of a movable entity within a predetermined area.

[0237] The system 60 further comprises a processing unit 62 adapted to process the real-time position information stream to generate occupancy information for each of a plurality of predetermined zones of a predetermined area.

[0238] The system 60 also includes an occupancy reason identification unit 63, which is suitable for processing the occupancy information using a machine learning model to determine one or more occupancy reasons 66, each occupancy reason indicating a reason why one or more predetermined areas are occupied by one or more movable entities, wherein the machine learning model defines a relationship between the occupancy information and the occupancy reasons.

[0239] The present invention may be a system, method and / or computer program product. The computer program product may include a computer-readable storage medium (or multiple media) having computer-readable program instructions thereon, for causing a processor to perform aspects of the present invention. Therefore, a computer-readable storage medium may be a tangible device that can retain and store instructions for use by an instruction execution device (e.g., a controller, a processor, or a processing system) to perform a method according to the present invention. The disclosed method is considered to be suitable for execution by a computer, and therefore may be a computer-implemented method.

[0240] The present invention is described herein with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It will be understood that each block of the flowchart illustration and / or block diagram and the combination of blocks in the flowchart illustration and / or block diagram can be implemented by computer-readable program instructions.

[0241] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functions and operations of possible implementations of the systems, methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram may represent a module, a fragment or a portion of an instruction, which includes one or more executable instructions for implementing (one or more) specified logical functions or a module for implementing a system of specified functions.

[0242] In some alternative implementations, the functions indicated in the blocks may not occur in the order indicated in the figures. For example, depending on the functionality involved, two blocks shown in succession may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. It should also be noted that each block of the block diagram and / or flow chart and combinations of blocks of the block diagram and / or flow chart may be implemented by a dedicated hardware-based system that performs the specified functions or actions or executes a combination of dedicated hardware and computer instructions.

[0243] By studying the drawings, the disclosure and the claims, those skilled in the art can understand and implement other variations of the disclosed embodiments when practicing the claimed invention. In the claims, the word "comprising" does not exclude other elements or steps, and the words "one" or "an" do not exclude multiple. A single processor or other unit can perform the functions of several items recorded in the claims. Although specific measures are recorded in mutually different dependent claims, this does not indicate that a combination of these measures cannot be used advantageously. The computer program can be stored / distributed on a suitable medium, such as an optical storage medium or solid-state medium provided with other hardware or as part of other hardware, but the computer program can also be distributed in other forms, such as distributed via the Internet or other wired or wireless telecommunications systems. Any figure mark in the claims should not be interpreted as limiting the scope.

Claims

1. A computer-implemented method for monitoring a movable entity within a predetermined area, the method comprising: Obtaining a real-time location RTL information stream identifying a current location of a movable entity within the predetermined area; processing the real-time location RTL information stream to generate occupancy information for each of a plurality of predetermined zones of the predetermined area, wherein the occupancy information indicates a number of movable entities of each of a plurality of entity types in each of the plurality of predetermined zones of the predetermined area; and The occupancy information is processed using a machine learning model to determine one or more occupancy reasons, each occupancy reason indicating the occurrence of a predetermined event within one or more of the plurality of predetermined zones of the predetermined area during a time when the real-time location RTL information stream is provided, wherein the machine learning model defines a relationship between the occupancy information and the occupancy reasons.

2. The computer-implemented method of claim 1 , wherein: The step of processing the real-time location RTL information stream comprises generating a sequence of occupancy data entries as the occupancy information, each occupancy data entry indicating a number of movable entities of each of the plurality of entity types within each of the plurality of predetermined zones of the predetermined area at a single point in time; and Each occupancy data entry is associated with a single point in time later than a previous occupancy data entry in the sequence of occupancy data entries such that the entire sequence indicates a number of movable entities of each of the plurality of entity types within each of the plurality of predetermined zones over a time period.

3. The computer-implemented method of claim 2, wherein: The machine learning model defines a relationship between a sequence of occupancy data entries and one or more occupancy reasons.

4. The computer-implemented method according to any one of claims 2 or 3, further comprising: determining one or more entity types for each at least one movable entity, wherein the step of generating the sequence of occupancy data entries comprises generating a sequence of occupancy data entries, wherein each occupancy data entry indicates a number of movable entities of each entity type within each of the plurality of predetermined zones of the predetermined area at a single point in time.

5. The computer-implemented method of claim 2, wherein: The occupancy information comprises a plurality of sequences of occupancy data entries, and the step of processing the occupancy information comprises using a machine learning model to individually process each sequence of occupancy data to determine one or more occupancy reasons.

6. The computer-implemented method of claim 5, wherein: The plurality of sequences form a series of sequences; and Each of the sequences in the series is associated with a time period starting at a later point in time than an immediately preceding sequence in the series of sequences.

7. The computer-implemented method of claim 6, wherein: Each sequence in the series of sequences is associated with a time period that overlaps with a time period associated with an immediately preceding sequence in the series of sequences.

8. A computer program product comprising code means for implementing the method according to any one of claims 1 to 7 when the program is run on a computer.

9. A system for monitoring a movable entity within a predetermined area, the system comprising: a stream receiving unit adapted to obtain a real-time location RTL information stream identifying a current location of a movable entity within said predetermined area; a processing unit adapted to process the real-time location RTL information stream to generate occupancy information for each of a plurality of predetermined zones of the predetermined area, the occupancy information indicating a number of movable entities of each of a plurality of entity types in each of the plurality of predetermined zones of the predetermined area; as well as An occupancy reason identification unit, which is suitable for processing the occupancy information using a machine learning model to determine one or more occupancy reasons, each occupancy reason indicating the occurrence of a predetermined event in one or more of the plurality of predetermined zones of the predetermined area during the time when the real-time location RTL information stream is provided, wherein the machine learning model defines a relationship between the occupancy information and the occupancy reasons.

10. The system according to claim 9, wherein: The processing unit is adapted to process the real-time location RTL information stream to generate a sequence of occupancy data entries, each occupancy data entry indicating a number of movable entities of each of the plurality of entity types within each of the plurality of predetermined zones of the predetermined area at a single point in time.

11. The system of claim 10, wherein: The processing unit is adapted to generate a plurality of sequences of occupancy data entries; and The occupancy reason identification unit is adapted to individually process each sequence of occupancy data of the plurality of sequences using a machine learning model to determine one or more occupancy reasons.

Images