Abstract: Human behavior is determined by sequential event detection by constructing a temporal-event graph with vertices representing primitive images of images of a video stream, and also of idle states associated with the respective primitive images. A human activity event is determined as a function of a shortest distance path of the temporal-event graph vertices.

Abstract: Multi-mode video event indexing includes determining a quality of object distinctiveness with respect to images from a video stream input. A high-quality analytic mode is selected from multiple modes and applied to video input images via a hardware device to determine object activity within the video input images if the determined level of detected quality of object distinctiveness meets a threshold level of quality, else a low-quality analytic mode is selected and applied to the video input images via a hardware device to determine object activity within the video input images, wherein the low-quality analytic mode is different from the high-quality analytic mode.

Abstract: Multi-mode video event indexing includes determining a quality of object distinctiveness with respect to images from a video stream input. A high-quality analytic mode is selected from multiple modes and applied to video input images via a hardware device to determine object activity within the video input images if the determined level of detected quality of object distinctiveness meets a threshold level of quality, else a low-quality analytic mode is selected and applied to the video input images via a hardware device to determine object activity within the video input images, wherein the low-quality analytic mode is different from the high-quality analytic mode.

Abstract: Techniques for creating a training technique for an individual are provided. The techniques include obtaining video of one or more events and information from a transaction log that corresponds to the one or more events, wherein the one or more events relate to one or more actions of an individual, classifying the one or more events into one or more event categories, comparing the one or more classified events with an enterprise best practices model to determine a degree of compliance, examining the one or more classified events to correct one or more misclassifications, if any, and revise the one or more event categories with the one or more corrected misclassifications, if any, and using the degree of compliance to create a training technique for the individual.

Abstract: Multiple event types are monitored for events, and surveillance data is stored for each event. Surveillance data for a primary event of one event type can be presented to a user, and surveillance data for a set of related events corresponding to another event type can be presented based on a set of relatedness criteria and the surveillance data for the primary event. A user can adjust the relatedness criteria to filter/adjust the surveillance data presented for the related event(s). A user interface can enable the user to simultaneously view the surveillance data for both events and adjust the relatedness criteria. In an illustrative application, the invention is utilized to detect fraudulent merchandise returns in a retail store.

Abstract: Human behavior is determined by sequential event detection by constructing a temporal-event graph with vertices representing adjacent first and second primitive images of a plurality of individual primitive images parsed from a video stream, and also of first and second idle states associated with the respective first and second primitive images. Constructing the graph is a function of an edge set between the adjacent first and second primitive images, and an edge weight set as a function of a discrepancy between computed visual features within regions of interest common to the adjacent first and second primitive images. A human activity event is determined as a function of a shortest distance path of the temporal-event graph vertices.

Abstract: Techniques for calibrating a classification system, wherein one or more objects in at least one video are classified, are provided. At least one view associated with the at least one video is obtained. The at least one view is partitioned into at least one region. A given object is classified in accordance with its location in reference to the at least one region. In an additional embodiment, one or more object models are obtained. At least one normalized size of the one or more objects is defined within at least one view associated with the at least one video in accordance with the one or more object models. The one or more objects are classified in accordance with the at least one defined normalized size.

Abstract: Multi-mode video event indexing includes determining a quality of object distinctiveness with respect to images from a video stream input. A high-quality analytic mode is selected from multiple modes and applied to video input images via a hardware device to determine object activity within the video input images if the determined level of detected quality of object distinctiveness meets a threshold level of quality, else a low-quality analytic mode is selected and applied to the video input images via a hardware device to determine object activity within the video input images, wherein the low-quality analytic mode is different from the high-quality analytic mode.

Abstract: Techniques for classifying one or more objects in at least one video, wherein the at least one video comprises a plurality of frames are provided. One or more objects in the plurality of frames are tracked. A level of deformation is computed for each of the one or more tracked objects in accordance with at least one change in a plurality of histograms of oriented gradients for a corresponding tracked object. Each of the one or more tracked objects is classified in accordance with the computed level of deformation.

Abstract: An approach that dynamically learns a set of attributes of an operator of a point of sale (POS) is provided. In one embodiment, there is an attribute tool, including an extraction component configured to receive sensor data of a set of moving objects, and extract a set of attributes from each of the set of moving objects captured within the scan area at the POS; an identification component configured to update an appearance model with the set of attributes from each of the set of moving objects; and an analysis component configured to analyze the appearance model to identify at least one of the set of moving objects as an operator of the POS.

Abstract: Techniques for calibrating a classification system, wherein one or more objects in at least one video are classified, are provided. At least one view associated with the at least one video is obtained. The at least one view is partitioned into at least one region. A given object is classified in accordance with its location in reference to the at least one region. In an additional embodiment, one or more object models are obtained. At least one normalized size of the one or more objects is defined within at least one view associated with the at least one video in accordance with the one or more object models. The one or more objects are classified in accordance with the at least one defined normalized size.

Abstract: Long-term understanding of background modeling includes determining first and second dimension gradient model derivatives of image brightness data of an image pixel along respective dimensions of two-dimensional, single channel image brightness data of a static image scene. The determined gradients are averaged with previous determined gradients of the image pixels, and with gradients of neighboring pixels as a function of their respective distances to the image pixel, the averaging generating averaged pixel gradient models for each of a plurality of pixels of the video image data of the static image scene that each have mean values and weight values. Background models for the static image scene are constructed as a function of the averaged pixel gradients and weights, wherein the background model pixels are represented by averaged pixel gradient models having similar orientation and magnitude and weights meeting a threshold weight requirement.

Abstract: Techniques for generating an alert based on absence of a given person in a transaction are provided. The techniques include monitoring, via video, a transaction, wherein the transaction includes presence of a given person in the transaction, relating the video of the transaction to a corresponding portion of a transaction log (TLOG), using the video and corresponding portion of the TLOG to detect if the given person in the transaction is present, and generating an alert if the given person is not present at the transaction.

Abstract: Techniques for automatically calibrating one or more regions of interest for video surveillance are provided. The techniques include at a user-defined frequency, determining if one or more regions of interest (ROIs) are present within a field of view of a camera, if one or more ROIs are present within the field of view of the camera, automatically calibrating the one or more ROIs within the field of view of the camera, and if one or more ROIs are not present within the field of view of the camera, sending an alert to a user.

Abstract: Objects within two-dimensional (2D) video data are modeled by three-dimensional (3D) models as a function of object type and motion through manually calibrating a 2D image to the three spatial dimensions of a 3D modeling cube. Calibrated 3D locations of an object in motion in the 2D image field of view of a video data input are computed and used to determine a heading direction of the object as a function of the camera calibration and determined movement between the computed 3D locations. The 2D object image is replaced in the video data input with an object-type 3D polygonal model having a projected bounding box that best matches a bounding box of an image blob, the model oriented in the determined heading direction. The bounding box of the replacing model is then scaled to fit the object image blob bounding box, and rendered with extracted image features.

Abstract: An approach to reduce the processing power required to monitor check-out stations. A subset of check-out stations may be monitored at any given time, instead of monitoring all check-out stations at all times. The subset of check-out stations is determined dynamically according to, but not limited to, cashier records, input parameters from the user, current lane activity, past lane activity, time of day, etc. Statistical models (e.g., effective population sampling and/or population hypothesis tests) are developed along these lines that guide the lane selection process, whereby increases in the false-negative rate due to failure to monitor particular lanes when events of interest occur are controlled. By monitoring fewer check-out stations, while maintaining target performance accuracy, the amount of data that end users must deal with is significantly reduced.

Abstract: Local models learned from anomaly detection are used to rank detected anomalies. The local models include image feature values extracted from an image field of video image data with respect to different predefined spatial and temporal local units, wherein anomaly results are determined by failures to fit to applied anomaly detection module local models. Image features values extracted from the image field local units associated with anomaly results are normalized, and image feature values extracted from the image field local units are clustered. Weights for anomaly results are learned as a function of the relations of the normalized extracted image feature values to the clustered image feature values. The normalized values are multiplied by the learned weights to generate ranking values to rank the anomalies.

Abstract: A method, data processing system, apparatus, and computer program product for monitoring objects. A plurality of images of an area is received. An object in the area is identified from the plurality of images. A plurality of points in a region within the area is identified from a first image in the plurality of images. The plurality of points has a fixed relationship with each other and the region. The object in the area is monitored to determine whether the object has entered the region. A determination that the object has not entered the region is made in response to identifying an absence of a number of the plurality of points in a second image in the plurality of images.

Abstract: An approach that allows for model based people counting is provided. In one embodiment, there is a generating tool configured to generate a set of person-shape models based on results of a cumulative training process; a detecting tool configured to detect persons in a camera field-of-view by using the set of person-shape models, and a counting tool configured to track detected persons upon crossing by the detected persons of a previously established virtual boundary.

Abstract: Training data object images are clustered as a function of motion direction attributes and resized from respective original into same aspect ratios. Motionlet detectors are learned for each of the sets from features extracted from the resized object blobs. A deformable sliding window is applied to detect an object blob in input by varying window size, shape or aspect ratio to conform to a shape of the detected input video object blob. A motion direction of an underlying image patch of the detected input video object blob is extracted and motionlet detectors selected and applied that have similar motion directions. An object is thus detected within the detected blob and semantic attributes of an underlying image patch extracted if a motionlet detectors fires, the extracted semantic attributes available for use for searching for the detected object.