WO2008127235A2 - Système de vision artificielle destiné à la gestion d'entreprise - Google Patents

Système de vision artificielle destiné à la gestion d'entreprise Download PDF

Info

Publication number
WO2008127235A2
WO2008127235A2 PCT/US2007/009095 US2007009095W WO2008127235A2 WO 2008127235 A2 WO2008127235 A2 WO 2008127235A2 US 2007009095 W US2007009095 W US 2007009095W WO 2008127235 A2 WO2008127235 A2 WO 2008127235A2
Authority
WO
WIPO (PCT)
Prior art keywords
sensor data
data
set forth
activity
enterprise
Prior art date
Application number
PCT/US2007/009095
Other languages
English (en)
Other versions
WO2008127235A3 (fr
Inventor
Andrew J. Griffis
Roger Karl Mikarl Undhagen
Tinku Acharya
Original Assignee
Avisere, Inc.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Avisere, Inc. filed Critical Avisere, Inc.
Priority to PCT/US2007/009095 priority Critical patent/WO2008127235A2/fr
Priority to CN200780053328.8A priority patent/CN101790717B/zh
Priority to EP07809015A priority patent/EP2193435A4/fr
Publication of WO2008127235A2 publication Critical patent/WO2008127235A2/fr
Publication of WO2008127235A3 publication Critical patent/WO2008127235A3/fr
Priority to HK11100917.7A priority patent/HK1146963A1/xx

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/06Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00Scenes; Scene-specific elements

Definitions

  • the invention relates to the field of machine (e.g., computer) vision and its application to the optimization and utilization of human activity and/or interaction within an enterprise (e.g., a business, neighborhood, home, other region or area of concern) to monitor events and aid in the automated or semi- automated decision making process to manage the enterprise efficiently and responsibly.
  • machine e.g., computer
  • enterprise e.g., a business, neighborhood, home, other region or area of concern
  • the invention disclosed, herein articulates a novel system design that uses sensor data (e.g., machine vision) events and combines them with key activity identifiers (e.g., specific human activity, animal activity, or interactions between humans, animals, machines, etc.) so that event data from sensors can be made available to decision logic for both real time (e.g., instant message, as an example, but not limited to) and delayed (e.g., database server, data mining application service, but not limited to) notification and/or recording of events and interrelationships of interest to a commercial, municipal type public enterprise or private enterprise, many of which enterprises also have a customer service oriented environment.
  • sensor data e.g., machine vision
  • key activity identifiers e.g., specific human activity, animal activity, or interactions between humans, animals, machines, etc.
  • system architecture of this invention enables the application of machine vision technology to automated or semi-automated enterprise situational awareness for deducing and optimizing the value (e.g., commercial value, but not limited to) of human interaction with other human subjects and/or animals (as these are commonly found in many enterprises and are known to be associated with human activity in some settings) and /or enterprise equipment in use by or associated with the human subjects.
  • value e.g., commercial value, but not limited to
  • an event is a phenomenon that occurs in the scene of particular interest viewed by the sensor that may be of particular interest to a particular user of the invention. Since the purpose of the invention is to provide situational awareness to a user having particular enterprise-related concerns, not all events result in a communication (e.g. message) to the user.
  • events are communicated to a user only insomuch as they are similar to events that the user determines are important, or that are deemed important by the invention through evidence of significance implied by the temporal and/or spatial behavior of event data (e.g., an event not easily confused with mere randomness in sensor data occurs at a time that is highly unusual, an event occurs repeatedly over time, or repeatedly in the same place, etc., to name a few examples).
  • event data e.g., an event not easily confused with mere randomness in sensor data occurs at a time that is highly unusual, an event occurs repeatedly over time, or repeatedly in the same place, etc., to name a few examples.
  • a key activity message that is useful in enterprise management will contain any or all of five elements: 1) what object is moving or changing (person, animal, vehicle, machine, other inanimate object) 2) what behavior the moving or changing object is engaged in, 3) where the object is located, 4) when the motion or change is occurring and 5) measurable interpretation of the change in phenomena or activity in order to make a decision by a machine or a human for a purpose.
  • Events that may be of interest to a retail store user would, for example, include the length of a line at the cash register exceeding three people (as an example), or some other number that the user deemed unacceptably high.
  • An example of an event of interest to a neighborhood organization would be a vehicle that appears in an alley several nights consecutively when people in the neighborhood are sleeping.
  • An example event that might be of interest for roadways and transportation concerns is the number of people in line waiting to cross a busy street at which a traffic signal has been placed.
  • the invention disclosed herein describes a system for use in managing activity of interest within an enterprise, comprising a computer configured to (i) receive sensor data that is related to key activity to the enterprise (such sensor data comprising data that is taken in by a sensor at a predetermined location irrespective of object compliance, and such key activity comprising an object, and the state of activity for the object with respect to the state of activity for other physically distinct objects in the sensor data), and (ii) process the sensor data to produce output that is related to key activity to the enterprise.
  • sensor data that is related to key activity to the enterprise
  • such sensor data comprising data that is taken in by a sensor at a predetermined location irrespective of object compliance, and such key activity comprising an object, and the state of activity for the object with respect to the state of activity for other physically distinct objects in the sensor data
  • process the sensor data to produce output that is related to key activity to the enterprise.
  • the computer is further configured to correlate combinations of past and present sensor data and key activity to the enterprise and the output includes feedback data based on the correlation.
  • the invention applies machine (e.g., computer) vision algorithms that use activity identifiers (e.g., data models, data records) applied to time series of sensor data (e.g., digital video image sequences on-line or off-line, different types of multimedia data) in order to detect events that have relevance to the user of the invention (e.g., the enterprise manager).
  • the invention applies decision logic, based on the knowledge discovery and information extraction through mining the available data using a computing system, regarding the statistical significance of these events to generate messages that can be sent immediately and/or stored for subsequent processing.
  • sensor data is quantitative information collected at a predetermined location and time at or near the enterprise, that can be used to deduce, using machine vision algorithms, events and key activities for the enterprise.
  • Sensor data is collected at the predetermined location irrespective of object compliance, meaning that the information about the object deduced by the invention is gathered without the participation of the object, in contrast to information that requires participation of the object for a typical biometric (e.g., fingerprint, face recognition, iris recognition, etc.) system.
  • biometric e.g., fingerprint, face recognition, iris recognition, etc.
  • an object is any physical entity, such entity being either animate or inanimate.
  • Animate objects are regarded as being either animals/creatures, humans, or machines.
  • Inanimate objects are regarded as being either man-made or natural.
  • Man-made objects are further regarded as being either machines, structures, furnishings or implements (e.g., tools, weapons, to name a few).
  • activity is motion in space and time such as would be observed by an object that is itself moving or is being moved, and includes the case for which no motion is occurring regarding an object.
  • an enterprise is a business, neighborhood, home, or other physical region or area of concern to one that is responsible for managing or monitoring activity within.
  • an event is a phenomenon that occurs in the scene of particular interest observed by the sensor that may be of particular interest to a particular user of the invention. Since the purpose of the invention is to provide situational awareness to a user having particular enterprise-related concerns, not all events result in a communication (e.g. message) to the user.
  • events are communicated to a user only insomuch as they are similar to events that the user determines are important, or that are deemed important by the invention through evidence of significance implied by the temporal and/or spatial behavior of event data (e.g., an event not easily confused with mere randomness in sensor data occurs at a time that is highly unusual, an event occurs repeatedly over time, or repeatedly in the same place, etc., to name a few examples).
  • a key activity is an event (or set of events) that the user determines is important, and/or that is deemed important by the invention through evidence of significance implied by the temporal and/or spatial behavior of event data.
  • An output comprising a key activity message that is useful in enterprise management will contain any or all of five elements: 1) what is moving or changing (person, animal, machine, other inanimate object) 2) what behavior the moving or changing object is engaged in, 3) where the object is located, 4) when the motion or change is occurring and 5) measurable interpretation of the change in phenomena or activity in order to make a decision by a machine or a human for a purpose.
  • Events that may be of interest to a retail store user would, for example, include the length of a line at the cash register exceeding three people, or some other number that the user deemed unacceptably high.
  • An example of an event of interest to a neighborhood organization would be a vehicle that appears in an alley several nights consecutively when people in the neighborhood are sleeping.
  • the event messages that are generated by the invention can be stored in a format (e.g., database) suitable for use with commercially available statistical analysis software.
  • the preferred embodiment inherently provides analysis of the recorded information for spatial and temporal correlation of events, signaling the enterprise principal(s) to examine these correlations and the underlying activity events when points of statistical significance are reached.
  • the system can then provide automatic updates to the key activity data of interest, thereby eliminating the need for an analyst/observer to review either the data produced by the system on ongoing basis, other than to gain an understanding of the underlying events that lead to messages.
  • the invention preferably uses a combination of embedded and application layer (e.g., in the context of an operating system and accessible by computer users) encryption or other security measures of the event data and raw data to safeguard the operation of the system and the integrity, of the components involved (e.g., to prevent commercial espionage of enterprise data).
  • embedded and application layer e.g., in the context of an operating system and accessible by computer users
  • Figure 1 shows the overall architecture of the invention and its primary components.
  • Data from a sensor is operated on by the invention, shown at the top of this diagram, and is reduced to messages as the data flow through the system.
  • These messages contain information of immediate and/or long term utility, and thereby are sent and/or used for subsequent long term processing, respectively.
  • Interfaces are provided to outside systems for both messaging and for database operations.
  • the user can input settings for the system to influence the processing, and can access data products from the database by the same means.
  • the invention exists to remove the necessity of ongoing data analysis in order to use the underlying algorithms, feedback is provided from the long-term analysis to the front end sensor data processing components.
  • Figure 2 illustrates one approach to the processing of sensor data in order to generate key activity event messages.
  • FIG. 3 shows the core elements of the processing system implied by the invention.
  • Figure 4 shows an example of the invention deployed as a client-server system such as might be deployed using conventional personal computer and server computer technology.
  • Figure 5 Shows the invention deployed in an embedded configuration, such as would be appropriate to a remote location that has limited network, power or security resources, or which benefits from embedded technology in some other way.
  • Figure 6 illustrates the modular design potential of the invention, which allows it to be deployed in many settings, such as those described in Figures 4 and 5.
  • a system of the present invention receives sensor data that is related to a key activity to the enterprise.
  • Figure 1 schematically illustrates the components and function of a system 100 according to the present invention.
  • data are input to a computer 101 and likewise output from the computer.
  • the invention functions by loading default key activity data 102 and/or having a user input key activity data 103 that then define the data analysis 104 that applies machine vision algorithms to input sensor data 105.
  • These algorithms produce outputs in the form of a key activity message 106 when statistically significant events corresponding to key activities are detected.
  • Such a key activity message can then be sent immediately or stored 107 for subsequent use.
  • the computer uses an interface provided to external electronic messaging systems 108.
  • the computer uses an interface to external data management systems 109.
  • algorithms are applied periodically to detect statistically significant temporal and/or spatial correlations 110 that can then lead to additional key activity (update) messages 111, complex/combined event messages 112, or such algorithms can provide feedback useful for altering the key activity data 113 through the key activity updating/identifying 114 function.
  • FIG. 2 schematically illustrates an example of the key activity analysis portion of the invention.
  • This algorithm example begins with the input of sensor data 105 to the computer 101.
  • the sensor data are then used to compute a sum of absolute differences across all sensor data points 201 between a prior point in time and the present time to detect statistically significant change; the sensor data points that are in the foreground (changing) can then be separated from those that are in the background (not changing), leading to foreground and background objects 202.
  • These objects are then localized by computing a centroid 203 and additional object calculations are made to determine object features and combine those objects that are supposed to be connected 204.
  • This algorithm compares current objects to historical (tracked) objects and determines whether the object is new or an updated version of an existing object 205.
  • a new track is initiated 206 and for existing objects track updates are made 206.
  • the tracker data having been updated, the statistical model used for tracking is then updated 207.
  • the current set of objects, locations and feature data are compared to key activity data 208 so that a key activity message can be generated 209 if a key activity is detected.
  • Figure 3 schematically illustrates the core elements of the system level elements of the invention.
  • the system is comprised of at least one sensor 301 that is connected via a sensor interface 302 to a computer 101 that contains memory 303 for storing sensor data and/or key activity messages.
  • the computer is also connected through a communication interface 304 to a communication network 305 that provides a means of communicating key activities of the enterprise to the enterprise manager.
  • the computer function can also be distributed to multiple computer components (e.g., workstations, servers or portable computing devices).
  • the computer 101 function is located in a single component.
  • FIG. 4 schematically illustrates an example of the invention deployed in a client-server configuration wherein some of the software functional blocks that could be implemented, using some common software and graphical user interface (GUI) objects, are shown.
  • GUI graphical user interface
  • This schematic designates the sensor data processing and message generation to the server (here shown as a web server); the GUI, data access, user configuration and message analysis/viewing capabilities are assigned to the client.
  • the server here shown as a web server
  • the GUI data access, user configuration and message analysis/viewing capabilities are assigned to the client.
  • all of the physical interfaces are integrated with the computer 101, and the functional blocks represent software objects that provide the computational interface to the physical interfaces; the software objects all run within the computer.
  • the sensor data 401 here shown also as optional stored data 407, are input to the computer 101, and is made available to a dynamic link library (DLL) object that performs the key activity analysis 402; the messages that result from the key activity analysis are generated/handled by a message handling/storing DLL 403 that further has access to a database 404 into which messages can be placed for storage and management.
  • DLL dynamic link library
  • a network interface object 405 provides a means of communication between the server and the client, and a configuration interface and executable 406 provides a means of allowing the client (user) to configure the invention.
  • the processes are managed and facilitated on the computer by a server/system interface (wrapper) 408.
  • the client side which is separated from the server side by a dashed line 409, is essentially a GUI application that runs on either the computer 101 or on a separate/equivalent networked computer.
  • the GUI client in this example of the invention consists of a main GUI window 410, a window for real-time data display 411, a window for setting user preferences 412 within the invention, a window for graphically viewing message data 413 such as a time series of event statistics, and finally the client GUI contains a configuration window 414 for the sensor and other key system components.
  • FIG. 5 schematically illustrates the invention deployed in an embedded (e.g., portable or similar) configuration in which the sensor, sensor data processing and external messaging interfaces are all integrated in a single module 501 that contains the computer 101 in which the sensor data are analyzed for occurrences of key activities, the sensor and corresponding sensor data 502 and message generation and communication components 503 that can use either a wired/wireless network connection 504 or a directly connected wired/wireless interface 505.
  • a workstation computer 506 provides the means of loading default key activity data .507 and can also perform computations to update key activity definitions 508, which can alternately be updated via user inputs 509.
  • the workstation computer can further analyze key activity message logs 510 by accessing a database 511 and the results can be communicated through an electronic messaging system interface 512 or an external data management system interface 513 so that a more distributed approach for the invention can be realized using one or more server 514 or server-like components.
  • some of the computer functions have been assigned/distributed to servers or workstation (both are computers) in the system.
  • Figure 6 schematically illustrates a system level embodiment of the invention deployed across a wide area network such as the internet 601.
  • the computer portion of the system occurs at multiple nodes on a network (meaning that the computing is distributed)
  • the sensor data and some key activity analysis pre-processing occur at one network node 602
  • the completion of sensor data key activity analysis occurs at second node 603
  • the resultant key activity messages are stored in a message log database 604 on a server 605 at a third node
  • the GUI for remote graphical analysis of key activity 606 is performed at a fourth network node.
  • This example illustrates the modular design of the invention so as to facilitate its use in a highly networked environment.
  • Sensor data is quantitative information collected at a predetermined location and time at or near the enterprise, that can be used to deduce, using machine vision algorithms, events and key activities for the enterprise.
  • Sensor data is collected at the predetermined location irrespective of object compliance, meaning that the information about the object deduced by the invention is gathered without the participation of the object, in contrast to information that requires participation of the object, such as would be the case for a typical biometric (e.g., fingerprint, face recognition, iris recognition, etc.) system.
  • biometric e.g., fingerprint, face recognition, iris recognition, etc.
  • An object is any physical entity, such entity being either animate or inanimate.
  • Animate objects are regarded as being either animals/creatures, humans, or machines.
  • Inanimate objects are regarded as being either man-made or natural.
  • Man-made objects are further regarded as being either machines, structures, furnishings or implements (e.g., tools, weapons, to name a few).
  • Activity is motion in space and time such as would be observed by an object that is itself moving or is being moved, and includes the case for which no motion is occurring regarding an object.
  • An enterprise is a business, neighborhood, home, or other physical region or area of concern to one that is responsible for managing or monitoring activity within.
  • An event is a phenomenon that occurs in the scene of particular interest observed by the sensor that may be of particular interest to a particular user of the invention.
  • a key activity is an event (or set of events) that the user determines is important, or that is deemed important by the invention through evidence of significance implied by the temporal and/or spatial behavior of event.
  • Key activity data are comprised of mathematical models that compactly represent human, vehicle, animal and inanimate object sensor data for events of interest to the enterprise.
  • the default key activity data are those key activity data installed with (or within) a system that are known or anticipated to have relevance to the user of the system in advance of system installation.
  • the purpose of these underlying models is to describe the default animate and inanimate objects that are of interest to the default user, in conjunction with the locations and times that are of particular interest for the specified objects.
  • These models will comprise of a set of general human features (e.g., body, limb, head, gait, etc.) and postures (e.g., standing, sitting, walking, running, jumping, lying, talking, waiving, writing, and other the patterns of movement), plus non-human but animate features and/or descriptors (e.g., canine, feline, aviary).
  • human body parts can be described mathematically in terms of connected ellipses (and other shapes) having areas in proportion to typical human proportions, and the way the ellipsoids (and other shapes) are connected in the sensor data can be used to determine a posture.
  • ellipsoids that have too small an area for a human shape and for which the connectivity of ellipsoids is more consistent with an animal than a human can be assigned an animate but non-human identifier.
  • the models that make up the key activity data can be constructed in many ways, including, but non limited to, the ellipsoidal human shape models discussed earlier.
  • Other mathematical descriptions of features in sensor data can include spatial (e.g., two-dimensional) histograms of foreground (e.g., moving) sensor regions of interest that indicate the number of distinct objects in a region; shape templates such as would be useful for matched-filter detection that allow for detecting objects in sensor data that are close to some reference shape (e.g., a matched filter that used a circular shape filter would be useful in detecting circular objects in sensor data); clustering and classification of objects by using feature vectors based on the sensor data (e.g., major and minor axes of ellipsoids as measures of aspect ratios); and frequency domain techniques such as the Fourier-Mellin transform, as an example, that can provide scale and rotation-invariant representations of sensor data that need to be represented compactly and compared to reference representations in order to assess how close a sensor data shape matches some known
  • An example of another useful mathematical description, or model, of shape in sensor data is a contour-based model that is compact and facilitates a small computational burden.
  • Many methods are available for this approach such as B-spline techniques for characterizing non-rigid shapes/features have been proven to be useful by some machine vision scientists.
  • a moving (foreground) object is detected in sensor data and a B-spline curve is fitted to the profile of the object in the sensor data, and this spline is used to describe the position and trajectory of the moving object, thereby enabling a complex description of an object that is tracked/observed with the invention, and yet using a minimum of sensor data to describe it (the B-spline having a relatively compact representation for a set of dense sensor data points).
  • the B-spline having a relatively compact representation for a set of dense sensor data points.
  • other machine vision models can be applied in the invention as well.
  • models that describe rigid, non-human objects such as vehicles can be used to correspondingly detect, describe, track and generally monitor the activity of such vehicles.
  • Human subjects are some of the more difficult to describe, and so are appropriate for this example discussion.
  • the same computer vision methods and skills used to approach the detection and measurement of human activities would be generally applicable to the the vehicular machine vision measurements.
  • the key activity data also include location, space and time data.
  • the default location data are simply the fields of view of the sensors that are installed as part of the system, so that, unless the user manually specifies otherwise, all sensor data (e.g., pixels) will be equally important or have the same meaning. However, the user may specify locations (e.g., bounding rectangles or other regions of interest) that have importance as a region that is to experience significant activity, little activity or none, for instance.
  • the default time data are simply the times of operation of the system, so that, unless the user manually specifies otherwise, all times will be equally important or have the same meaning when a system is first installed.
  • the user can manually specify periods of particular interest for either high activity, low activity or none (e.g., a back entrance of a restaurant should be inactive during non-business hours, and events during those hours would be important at even single event levels).
  • the default key activity data will include classes of activity that combine both animate and inanimate data models and constitute a composite activity.
  • the combination of walking events occurring within a doorway object would be class of event that is of interest to most users and therefore would be an appropriate default to include.
  • the observation of the number of pedestrians waiting to cross a street, combined with the state of activity of vehicles (traffic) would represent such a combination of activities.
  • the direction of movement may have significance where the direction of traffic through a one-way roadway has safety implications.
  • the key activity data also include location and time data.
  • the default location data are simply the fields of view of the sensors that are installed as part of the system, so that, unless the user manually specifies otherwise, all sensor data (e.g., pixels) will be equally important or have the same meaning. However, the user may specify locations (e.g., bounding rectangles or other regions of interest) that have importance as a region that is to experience significant activity, little activity or none, for instance.
  • the key activity data thus described are best implemented as a data structure that is read into the system, though is it possible to conceive of other means of storing, retrieving and manipulating such data (e.g., as multiple single data type records, simple buffers, but not limited to).
  • This element of the invention provides a means of adding to the default key activity data either automatically (i.e., based on the data that are collected and analyzed subsequent to using the invention in deployed system) or manually.
  • the manual addition of key activity data for the invention to monitor is achieved using an interactive user interface (e.g., graphical user interface, voice portal, or GUI) or its equivalent in the case where a graphical interface is not possible (e.g., a menu on a cell phone, computer monitor or dial-in phone portal).
  • GUI graphical user interface
  • the automatic addition of data to the list of key activities is achieved by statistical inference from the analysis of the message log that accumulates as the system is operated.
  • Isolated events that occur frequently e.g., person walking through doorway
  • clustered events that occur frequently e.g., group of people standing in line at the cash register
  • isolated events that occur infrequently e.g., person loitering by the back entrance to the restaurant late at night after business hours
  • the invention already includes provision for default key activity data that are appropriate to the industry or setting in which it is deployed. However, it is anticipated that there will be particular activities, objects or classes of objects and/or events that will not be included in the default activities, yet are important to the user. Thus, the system includes the ability to select additional data elements from a pre-defined list, or to add custom data elements that the user obtains through independent means.
  • GUI graphical user interface
  • the GUI might use a list of selected check-boxes that are initialized to the default settings, but include unselected boxes that can be used instead of or in addition to the defaults.
  • the GUI in order to facilitate the inclusion of additional model data, would have a button or other control that enables the user to load model data that are compatible with the particular data structure in use.
  • the user can input the locations of system components for use in spatial correlation and the location in the sensor data of objects of particular interest, even if the user is not able or interested in specifying a model for objects thus specified.
  • this selection of objects can be as simple as drawing an enclosing rectangle or other shape with a point device such as a mouse.
  • a point device such as a mouse.
  • Such data can also be entered through a less sophisticated interface (e.g., a voice portal) but the entry of such data becomes cumbersome, so that the GUI is again the preferred embodiment.
  • the invention uses sensor data (e.g., digital video data, infrared data, microwave data, other multimedia data including but not limited to audio) from user selected sensors that provide information about the enterprise of concern.
  • sensor data e.g., digital video data, infrared data, microwave data, other multimedia data including but not limited to audio
  • the sensor data can originate from a single sensor or multiple sensors, and can be co-located with other parts of the invention or can originate in remote, networked locations.
  • the sensor data by virtue of either a unique connection (e.g., data port) on the sensor data analysis computer resource (e.g., server or desktop computer) or by virtue of a unique sensor identifier (e.g., a sensor serial number or watermark placed on its data prior to transmission of the data to the sensor data analysis computer resource) must be associated with a particular location, as the location of events derived from sensor data is an important component of the key activity data.
  • a unique connection e.g., data port
  • a unique sensor identifier e.g., a sensor serial number or watermark placed on its data prior to transmission of the data to the sensor data analysis computer resource
  • Sensor data is collected at the predetermined location irrespective of object compliance, meaning that the information about the object deduced by the invention is gathered without the participation of the object, in contrast to information that requires participation of the object, such as would be the case for a typical biometric (e.g., fingerprint, face recognition, iris recognition, etc.) system.
  • a biometric e.g., fingerprint, face recognition, iris recognition, etc.
  • a sensor were placed so as to observe the activities of a person roaming the aisles of a retail enterprise, significant information about the human object features could be measured that accurately describe the human object without the knowledge of the object, let alone the participation of the object; participation would require the object to pose, to somehow collaborate with the measurement, or otherwise facilitate the gathering of the activity information.
  • This aspect of the invention involves the application of machine vision and machine intelligence algorithms to the sensor data (e.g., digital video image sequences or individual images) including, but not limited to the following algorithm elements:
  • Localizing and tracking of foreground objects involves computing the location of objects as indicated by the placement in the sensor data (e.g., the centroid of a grouping of sensor pixels in a digital image, to name one possible approach).
  • Tracking involves the measurement of location, the measurement of the statistics (e.g., mean, variance, covariance) of motion, including but not limited to velocity and acceleration. Tracking can be used to maintain information in the past location of an object, the current location of an object and also to predict future locations.
  • a very simple system to detect and track a human in a sensor data sequence might involve: computing differences between sequential sensor data (e.g., sequential images in a video sequence) that indicate sensor data that are changing enough to indicate motion; filtering these differences (e.g., applying a Gaussian matched filter to smooth the difference image) to reduce susceptibility to random events and thresholding the filtered data to yield sensor data regions that are undergoing significant change; identifying contours in the thresholded data and extracting the locations of the contour points; using the contour points to describe the area enclosed within contour(s) and also determining the major axes of the contour(s); deciding based on the contour(s) thus calculated whether the shape could be a human; if human-like, computing a centroid of the contour(s) that can be used to track the contoured object (understood at this point to be human-like) from one sensor data set to the next in time by measuring the centroid of the object over time and using this history to both form an accurate estimate
  • the result of the application of the analysis algorithm to the sensor data is an output event and event and/or key activity data record that contains data that may contain, but are not limited to, the specific nature of the event, the event time, position, type and spatial and temporal nature of the object. These will vary somewhat with the particular algorithm chosen to detect the objects, but could include, for example, the velocity, vector gradient data such as the divergence and curl, and spectral behavior of the object if such data are available from the sensor.
  • the type of object closely matches that of a key activity/object model
  • this type identity will be declared as part of the output; however, it is envisioned that, from time to time, the identity of an object will not be known and will be left as an unknown, to be determined later, or that the event is simply unknown but, owing to availability of message storage capacity, is submitted for message generation in order to facilitate the discovering of new key activities based on the analysis of event and message data over time.
  • the output data record may also include the statistical certainty of the analysis data products, e.g., the likelihood ratio in the case of the preferred embodiment that utilizes statistical hypothesis testing to detect and classify objects.
  • the output data record(s) from the analysis algorithm can be used to generate messages that represent a decision to store the event as declared.
  • the decision as to whether or not to generate a message, or message generation logic is made based on the statistical significance of events and may include the preferences of the user as to the level of statistical significance that is required before a message is generated (e.g., the user may specify that events must have a probability of false alarm of less than 1 per hour, which would then directly affect the threshold setting for the test statistic used in the likelihood test, in the case for which that particular class of test is used).
  • the data produced by the sensor data analysis algorithm are formatted for use by both the data storage (e.g., database or other organized means of storage) and messaging modules (e.g., email, instant message, or other means of communicating the message) and sent to one or both of these modules.
  • data storage e.g., database or other organized means of storage
  • messaging modules e.g., email, instant message, or other means of communicating the message
  • the preferred embodiment involves the storage of the message in a known location so that additional processing can be performed on the stored messages (the message log) to deduce more complex, long-term activities across many activities and locations being observed.
  • the preferred embodiment also involves the storage of events that do not necessarily match a known key activity.
  • the storage as messages of these events that are in some way unknown or otherwise regarded as insignificant provides a means of automatic or semi-automatic learning by the user about the events that take place in the enterprise of interest.
  • Such event messages could thus be stored, with appropriate descriptors in the stored data to indicate that the event is in some way unknown or incompletely determined.
  • Encryption is used in the invention as a means of protecting the sensor data and/or messages that are generated from observation by those outside the base of intended users. Since it is clear that the invention could be used in situations where confidential events are observed (e.g., closing the bank till, locking or unlocking the entry to a retail establishment) the use of encryption to protect data and/or message traffic is important for providing a high level of customer service. Encryption would occur in any deployment of the technology at the point where the data and/or key event messages are generated. Once the data and/or message is at a trusted party on the other side of the network, decryption can be implemented, thereby providing a secure data pipe for the transaction. There are many possible methods of providing the encryption, such as that currently employed by the secure socket layer (SSL) protocol that has found widespread use on the internet, to name one common example.
  • SSL secure socket layer
  • the messages received from the message generation component of the invention will be stored in a predetermined location or locations (e.g., local memory, local or remote database on mass storage media) and/or placed in a message queue for a messaging system having an interface to public or proprietary messaging system.
  • a predetermined location or locations e.g., local memory, local or remote database on mass storage media
  • this element of the invention is essentially an interface to databases and message handlers.
  • the preferred embodiment will encompass both a private database that facilitates the ongoing analysis of message data, an external commercially-available database (e.g., SQL or equivalent) for occasional on- demand access by the user and/or user-approved analysts, and a commercial message proxy service that allows a single interface to multiple instant message (e.g., AOL) and simple message (e.g., Vodaphone) standards, thereby removing the need to incorporate such standards into the system.
  • a private database that facilitates the ongoing analysis of message data
  • an external commercially-available database e.g., SQL or equivalent
  • a commercial message proxy service that allows a single interface to multiple instant message (e.g., AOL) and simple message (e.g., Vodaphone) standards, thereby removing the need to incorporate such standards into the system.
  • This element of the invention is essentially a queue for messages that accommodate a commercial or proprietary standard for messaging, including email, instant messaging and SMS (simple messaging system) data such as is used for mobile phones. It is envisioned that the invention may need to send proprietary messages to non-public resources (e.g., a microcontroller on a private or public network that can receive such messages and controls hardware that needs to be triggered by the information generated with the invention). However, the preferred embodiment uses a standard messaging interface/method so that system complexity and instability for the user is minimized.
  • This element of the invention is essentially a data storage (e.g., database) interface for writing message data records.
  • data storage e.g., database
  • non-standard proprietary interfaces may be necessary, but the preferred embodiment relies on commercially available standards to that the user experience is as simple and uncomplicated as possible, given the overall goal of the invention to provide advanced technology to those not having machine vision or other advance technology expertise.
  • the purpose of this element of the invention is to analyze the stored event and key activity messages for statistically significant combined, or complex, activities (e.g., events that are comprised of multiple messages separated by time or location) and multiple event cross-correlations (e.g., events that happen in the same region of interest, at the same time in the same region of interest, or at the same time in different regions of interest).
  • complex activities e.g., events that are comprised of multiple messages separated by time or location
  • multiple event cross-correlations e.g., events that happen in the same region of interest, at the same time in the same region of interest, or at the same time in different regions of interest.
  • the invention allows for multiple events (i.e., unknown or incompletely determined) and/or key activities to be analyzed for the occurrence of more complex key activity than can be detected by the aforementioned analysis of sensor data for key activity.
  • the output of such correlation can be fed back to the data structure that identifies key activity to the enterprise, to enable modification of the data structure that identifies key activity to the enterprise.
  • this element can be implemented in many ways, given the intense activity in such data mining and related technology in recent years.
  • the preferred embodiment again is to rely upon robust statistical hypothesis testing methods that allow for characterization of multiple hypotheses and that do not necessarily rely upon a priori specification of event statistics (e.g., a normal distribution is not generally appropriate for detecting and tracking events that lie well away from a prominent mode of a probability density function).
  • the invention can form complex event messages that the user specifies or that become apparent through the message update mechanism (see prior paragraph). These complex event messages are handled the same as the simpler key activity messages, once generated.
  • Figures 2-5 Examples of possible embodiments of the major features delineated in Figure 1 are provided in Figures 2-5, which are described in the following, in relation to Figure 1.
  • Figure 2 shows an example of a simplistic way that sensor data (assumed to be digital video, or sequences of digital images or frames) can be processed in order to generate key activity event messages. This is a simplistic example of some of the computations described in Figure 1 as "Analyze Sensor Data for Occurrence of Key Activities" and also "Generate a Key Activity Message.”
  • the foreground data can be analyzed to localize, or compute the position of, foreground objects. Subsequently, the foreground objects are analyzed for features, and if multiple objects are detected and found to be part of a larger whole (e.g., connected component analysis), these components can be connected/associated and thereby associated with the appropriate objects.
  • the tracking algorithm (“tracker”; a tracking algorithm simply keeps track of where an object is, and is typically capable of predicting future object locations based on prior object location history) is applied and the detected objects can be compared to existing tracked objects so that new objects can begin to be tracked and/or existing objects can have their tracks updated.
  • the statistical model used for the tracker can be updated; this generally involves updating the error co variance for the tracked objects, updating the stored current locations and updating the predicted track for the next instance of sensor data.
  • the features and location of the object can all be compared to the key activity events of interest in order to make a determination as to whether a key activity event has occurred. If such an event has occurred, a message is generated.
  • Figure 3 shows the core elements of the processing system for practicing the invention.
  • the system must have a processor (e.g., a microcomputer, digital signal processor, or application specific integrated circuit); some storage or memory that allows for computer programs to be run, sensor data to be manipulated and stored and also to enable the storage of messages generated from the short- or long-term analysis of sensor data and key activity data; interfaces for both communicating the outputs of the invention and for the acquisition of the sensor data.
  • a processor e.g., a microcomputer, digital signal processor, or application specific integrated circuit
  • some storage or memory that allows for computer programs to be run, sensor data to be manipulated and stored and also to enable the storage of messages generated from the short- or long-term analysis of sensor data and key activity data
  • interfaces for both communicating the outputs of the invention and for the acquisition of the sensor data.
  • Figure 4 shows an example of the invention deployed as a client-server application that would run on conventional client (e.g., desktop or mobile computer) and server (e.g., web server, local area network server such as found in a typical enterprise setting).
  • client e.g., desktop or mobile computer
  • server e.g., web server, local area network server such as found in a typical enterprise setting.
  • this figure shows how some of the elements of Figure 1 can be assigned to computer executable modules, shown here for example as dynamic link library (DLL) modules such as would be commonly found on a modern computer.
  • DLL dynamic link library
  • the figure is separated into two halves by a dashed line.
  • the upper portion is the server part and the lower portion is the client part.
  • the server contains a DLL that handles the key activity analysis and also one for handling the generation and handling/storing of messages.
  • the server side in this example also has the interface to the sensor and/or stored datafiles (e.g., AVI), the data from which are fed to the analysis DLL.
  • the server provides the means of controlling and maintaining the configuration of the system (sensors, events of interest, other system particulars) through its configuration interface.
  • the server also provides a network interface for the executable components, which are all accessible through a interface module (sometimes called a wrapper) and the network module (e.g., TCP, or transport control protocol).
  • the server provides the interface to the database that contains the message data archive, and which, through the database DLL can perform the necessary computations to deduce statistical significance of complex events and of new/recurring events logged in the database.
  • the client side illustrates the typical range of features that would be expected if the invention were accessed by a user from the client side through a typical desktop computer operating system interface (e.g., Microsoft Windows), and where tools relating to analysis, settings, configurations, sensor data viewing and the like are all provided.
  • Windows or interfaces are provided to the user for configuring the system (selecting/determining key activities), for setting preferences, for viewing results from the database DLL and for viewing a variety of data generated through the system.
  • Figure 5 Shows the invention deployed in an embedded configuration, which occurs when it is necessary to reduce the raw sensor data (a form of compression) prior to transmitting the data across a network, where both security and bandwidth may be significantly lower than in a setting where all the system components are contained within a single building/structure).
  • ASIC application specific integrated circuit
  • general purpose programmable processor e.g., microprocessor, digital signal processor
  • FIG. 1 the major elements of Figure 1 are all visible. However, they are distributed across a network and some (the analysis of activities and the message generation) are embedded in a module that contains the sensor that is providing the data to analyze.
  • a direct connection from an outside wireless or wired device is also shown here, since it is envisioned that direct messaging from such an embedded module to an outside device (e.g.,. held by a person for whom the messages are intended) will be very convenient to a user that is, for example, in a remote location without adequate network infrastructure.
  • the invention is structured so that the sensor data, the sensor data processing, the message database and the message data processing and access can all take place on separate, networked resources.
  • the sensor lower left corner
  • the sensor is deployed remote to the rest of the system but having a network connection for the sending of either raw sensor data or reduced data (via an ASIC for example).
  • the sensor data herein are transmitted to the sensor data analysis computer shown on the upper right, which generates messages that are logged by yet another server in a message log database (lower right).
  • This (lower right) server also can run database and statistical applications to serve up data results of interest to a user of the invention.
  • the user can make requests for such results from a remote, but networked, location via a portable computer (e.g., laptop), as shown in the upper left.
  • a portable computer e.g., laptop

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • Theoretical Computer Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • Game Theory and Decision Science (AREA)
  • General Business, Economics & Management (AREA)
  • Development Economics (AREA)
  • Educational Administration (AREA)
  • Multimedia (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Testing Or Calibration Of Command Recording Devices (AREA)

Abstract

L'invention concerne un système de gestion d'une activité particulière dans une entreprise. Le système comprend un ordinateur configuré afin de (i) recevoir des données de capteur qui sont liées à une activité principale de l'entreprise, ladite activité principale comprenant un type d'objet et l'activité de l'objet à un endroit prédéterminé associé à l'entreprise, le capteur fournissant des informations à partir desquelles un type et une activité d'objet à l'endroit prédéterminé peuvent être dérivés, (ii) traiter les données de capteur afin de produire un résultat qui est lié à l'activité principale de l'entreprise, et (ii) stocker les informations extraites des données traitées de manière adéquate en vue d'une extraction de connaissances et d'une analyse future. Selon un mode de réalisation préféré, l'objet est un humain, une machine ou un véhicule, et l'ordinateur est en outre configuré afin de corréler les données de capteur avec l'activité principale de l'entreprise et le résultat comprend des données de retour basées sur la corrélation.
PCT/US2007/009095 2007-04-13 2007-04-13 Système de vision artificielle destiné à la gestion d'entreprise WO2008127235A2 (fr)

Priority Applications (4)

Application Number Priority Date Filing Date Title
PCT/US2007/009095 WO2008127235A2 (fr) 2007-04-13 2007-04-13 Système de vision artificielle destiné à la gestion d'entreprise
CN200780053328.8A CN101790717B (zh) 2007-04-13 2007-04-13 用于企业管理的机器视觉系统
EP07809015A EP2193435A4 (fr) 2007-04-13 2007-04-13 Système de vision artificielle destiné à la gestion d'entreprise
HK11100917.7A HK1146963A1 (en) 2007-04-13 2011-01-28 Machine vision system for enterprise management

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/US2007/009095 WO2008127235A2 (fr) 2007-04-13 2007-04-13 Système de vision artificielle destiné à la gestion d'entreprise

Publications (2)

Publication Number Publication Date
WO2008127235A2 true WO2008127235A2 (fr) 2008-10-23
WO2008127235A3 WO2008127235A3 (fr) 2009-02-19

Family

ID=39864493

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2007/009095 WO2008127235A2 (fr) 2007-04-13 2007-04-13 Système de vision artificielle destiné à la gestion d'entreprise

Country Status (4)

Country Link
EP (1) EP2193435A4 (fr)
CN (1) CN101790717B (fr)
HK (1) HK1146963A1 (fr)
WO (1) WO2008127235A2 (fr)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8893163B2 (en) 2012-05-14 2014-11-18 Cihan Atkin Method and system for viewing movies
US9525911B2 (en) 2014-03-27 2016-12-20 Xcinex Corporation Techniques for viewing movies

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5369252B2 (ja) * 2011-03-23 2013-12-18 株式会社日立製作所 計算機システム及び計算機システムを用いた情報提示方法
CA2900765A1 (fr) * 2013-02-08 2014-08-14 Emotient Recolte de donnees de formation d'apprentissage machine pour une reconnaissance d'expression
US9256786B2 (en) * 2013-10-15 2016-02-09 Ge Aviation Systems Llc Method of identification from a spatial and spectral object model
US20160260024A1 (en) * 2015-03-04 2016-09-08 Qualcomm Incorporated System of distributed planning

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5097328A (en) * 1990-10-16 1992-03-17 Boyette Robert B Apparatus and a method for sensing events from a remote location
US6806808B1 (en) * 1999-02-26 2004-10-19 Sri International Wireless event-recording device with identification codes
US6961000B2 (en) * 2001-07-05 2005-11-01 Amerasia International Technology, Inc. Smart tag data encoding method
WO2003007121A2 (fr) * 2001-07-12 2003-01-23 Atrua Technologies, Inc. Procede et systeme permettant de determiner la confidence dans une transaction numerique
US20030040925A1 (en) * 2001-08-22 2003-02-27 Koninklijke Philips Electronics N.V. Vision-based method and apparatus for detecting fraudulent events in a retail environment
US7688349B2 (en) * 2001-12-07 2010-03-30 International Business Machines Corporation Method of detecting and tracking groups of people
US7019639B2 (en) * 2003-02-03 2006-03-28 Ingrid, Inc. RFID based security network
US20040260513A1 (en) * 2003-02-26 2004-12-23 Fitzpatrick Kerien W. Real-time prediction and management of food product demand
US20040236188A1 (en) * 2003-05-19 2004-11-25 Ge Medical Systems Information Method and apparatus for monitoring using a mathematical model
US20050091093A1 (en) * 2003-10-24 2005-04-28 Inernational Business Machines Corporation End-to-end business process solution creation

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
None

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8893163B2 (en) 2012-05-14 2014-11-18 Cihan Atkin Method and system for viewing movies
US9525911B2 (en) 2014-03-27 2016-12-20 Xcinex Corporation Techniques for viewing movies

Also Published As

Publication number Publication date
CN101790717B (zh) 2014-07-16
HK1146963A1 (en) 2011-07-22
WO2008127235A3 (fr) 2009-02-19
CN101790717A (zh) 2010-07-28
EP2193435A2 (fr) 2010-06-09
EP2193435A4 (fr) 2012-07-11

Similar Documents

Publication Publication Date Title
US7617167B2 (en) Machine vision system for enterprise management
US11196966B2 (en) Identifying and locating objects by associating video data of the objects with signals identifying wireless devices belonging to the objects
US10049288B2 (en) Managed notification system
CN110428522A (zh) 一种智慧新城的智能安防系统
US11115629B1 (en) Confirming package delivery using audio/video recording and communication devices
US10593174B1 (en) Automatic setup mode after disconnect from a network
US11018939B1 (en) Determining product compatibility and demand
US11240474B1 (en) Reporting connectivity problems for electronic devices
US11217076B1 (en) Camera tampering detection based on audio and video
US11257226B1 (en) Low-overhead motion classification
CN102099813A (zh) 被管理的基于生物特征识别的通知系统和方法
CN108898104A (zh) 一种物品识别方法、装置、系统及计算机存储介质
EP2193435A2 (fr) Système de vision artificielle destiné à la gestion d'entreprise
US10742939B1 (en) Security video data processing systems and methods
CN111757069B (zh) 一种基于智能门铃的监控防盗方法及装置
US10733857B1 (en) Automatic alteration of the storage duration of a video
CN109495723A (zh) 一种基于雾计算的家用智能监控系统
WO2017058587A1 (fr) Procédé et système de vérification d'image
Legrand et al. Study of autoencoder neural networks for anomaly detection in connected buildings
US10943442B1 (en) Customized notifications based on device characteristics
CN114972727A (zh) 用于多模态神经符号场景理解的系统和方法
US20210142084A1 (en) Managed notification system
Bushra et al. Predicting Anomalous and Consigning Apprise During Heists
US10914811B1 (en) Locating a source of a sound using microphones and radio frequency communication
US20220335725A1 (en) Monitoring presence or absence of an object using local region matching

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 200780053328.8

Country of ref document: CN

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 07809015

Country of ref document: EP

Kind code of ref document: A2

NENP Non-entry into the national phase

Ref country code: DE

WWE Wipo information: entry into national phase

Ref document number: 2007809015

Country of ref document: EP