Abstract: Methods and apparatus are disclosed for interpreting a deep neural network (DNN) using a Semantic Coherence Analysis (SCA)-based interpretation technique. In embodiments, a multi-layered DNN that was trained for one task is analyzed using the SCA technique to select one layer in the DNN that produces salient features for another task. In embodiments, the DNN layers are tested with test samples labeled with a set of concept labels. The output features of a DNN layer are gathered and analyzed according to the concepts. In embodiments, the output is scored with a semantic coherence score, which indicates how well the layer separates the concepts, and one layer is selected from the DNN based on its semantic coherence score. In some embodiments, a support vector machine (SVM) or additional neural network may be added to the selected layer and trained to generate classification results based on the outputs of the selected layer.

Abstract: Alias capture to support searching for an object-of-interest is disclosed. A method includes capturing, using a camera with a defined field of view, video image frames that include a moving object-of-interest. The method also includes tracking the object-of-interest over a period of time starting when the object-of-interest enters the field of view and ending when the object-of-interest exits the field of view. The method also includes detecting, at a point in time in-between the start and end of the period of time of the tracking, a threshold exceeding change in an appearance of the object-of-interest. The method also includes creating, before the end of the period of time of the tracking, a new object profile for the object-of-interest in response to the detecting of the threshold exceeding change.

Abstract: Methods, systems, and techniques for monitoring an object-of-interest within a region involve receiving at least data from two sources monitoring a region and correlating that data to determine that an object-of-interest depicted or represented in data from one of the sources is the same object-of-interest depicted or represented in data from the other source. Metadata identifying that the object-of-interest from the two sources is the same object-of-interest is then stored for later use in, for example, object tracking.

Abstract: There is provided an appearance search system comprising one or more cameras configured to capture video of a scene, the video having images of objects. The system comprises one or more processors and memory comprising computer program code stored on the memory and configured when executed by the one or more processors to cause the one or more processors to perform a method. The method comprises identifying one or more of the objects within the images of the objects. The method further comprises implementing a learning machine configured to generate signatures of the identified objects and generate a signature of an object of interest. The system further comprises a network configured to send the images of the objects from the camera to the one or more processors.

Abstract: Alias capture to support searching for an object-of-interest is disclosed. A method includes capturing, using a camera with a defined field of view, video image frames that include a moving object-of-interest. The method also includes tracking the object-of-interest over a period of time starting when the object-of-interest enters the field of view and ending when the object-of-interest exits the field of view. The method also includes detecting, at a point in time in-between the start and end of the period of time of the tracking, a threshold exceeding change in an appearance of the object-of-interest. The method also includes creating, before the end of the period of time of the tracking, a new object profile for the object-of-interest in response to the detecting of the threshold exceeding change.

Abstract: Alias capture to support searching for an object-of-interest is disclosed. A method includes capturing, using a camera with a defined field of view, video image frames that include a moving object-of-interest. The method also includes tracking the object-of-interest over a period of time starting when the object-of-interest enters the field of view and ending when the object-of-interest exits the field of view. The method also includes detecting, at a point in time in-between the start and end of the period of time of the tracking, a threshold exceeding change in an appearance of the object-of-interest. The method also includes creating, before the end of the period of time of the tracking, a new object profile for the object-of-interest in response to the detecting of the threshold exceeding change.

Abstract: Methods and apparatus are disclosed for interpreting a deep neural network (DNN) using a Semantic Coherence Analysis (SCA)-based interpretation technique. In embodiments, a multi-layered DNN that was trained for one task is analyzed using the SCA technique to select one layer in the DNN that produces salient features for another task. In embodiments, the DNN layers are tested with test samples labeled with a set of concept labels. The output features of a DNN layer are gathered and analyzed according to the concepts. In embodiments, the output is scored with a semantic coherence score, which indicates how well the layer separates the concepts, and one layer is selected from the DNN based on its semantic coherence score. In some embodiments, a support vector machine (SVM) or additional neural network may be added to the selected layer and trained to generate classification results based on the outputs of the selected layer.

Abstract: There are described methods and systems for facilitating identification of an object-of-interest. A face similarity score and a body similarity score of a query image relative to a gallery image are determined. A fused similarity score of the query image relative to the gallery image is determined by applying a relationship between the face similarity score, the body similarity score, and the fused similarity score. The fused similarity score is indicative of whether or not the object-of-interest and the potential object-of-interest are the same object-of-interest. For example, a machine learning process is used to fuse the face similarity score and the body similarity into the fused similarity score. The process is repeated for multiple gallery images. The gallery images may then be ranked according to their respective fused similarity scores.

Abstract: Methods, systems, and techniques for monitoring an object-of-interest within a region involve receiving at least data from two sources monitoring a region and correlating that data to determine that an object-of-interest depicted or represented in data from one of the sources is the same object-of-interest depicted or represented in data from the other source. Metadata identifying that the object-of-interest from the two sources is the same object-of-interest is then stored for later use in, for example, object tracking.

Abstract: Methods, systems, and techniques for interfacing with a user to facilitate an image search for a person-of-interest. A face thumbnail of the person-of-interest, a body thumbnail of the person-of-interest, and image search results of the person-of-interest are concurrently displayed on a display to help a user identify the person-of-interest who appears in one or more of a collection of video recordings. The user may provide feedback to the system regarding whether image search results show the person-of-interest, which feedback may be used to refine the image search results.

Abstract: There is provided an appearance search system comprising one or more cameras configured to capture video of a scene, the video having images of objects. The system comprises one or more processors and memory comprising computer program code stored on the memory and configured when executed by the one or more processors to cause the one or more processors to perform a method. The method comprises identifying one or more of the objects within the images of the objects. The method further comprises implementing a learning machine configured to generate signatures of the identified objects and generate a signature of an object of interest. The system further comprises a network configured to send the images of the objects from the camera to the one or more processors.

Abstract: Methods, systems, and techniques for monitoring an object-of-interest within a region involve receiving at least data from two sources monitoring a region and correlating that data to determine that an object-of-interest depicted or represented in data from one of the sources is the same object-of-interest depicted or represented in data from the other source. Metadata identifying that the object-of-interest from the two sources is the same object-of-interest is then stored for later use in, for example, object tracking.

Abstract: There is provided an appearance search system comprising one or more cameras configured to capture video of a scene, the video having images of objects. The system comprises one or more processors and memory comprising computer program code stored on the memory and configured when executed by the one or more processors to cause the one or more processors to perform a method. The method comprises identifying one or more of the objects within the images of the objects. The method further comprises implementing a learning machine configured to generate signatures of the identified objects and generate a signature of an object of interest. The system further comprises a network configured to send the images of the objects from the camera to the one or more processors.

Abstract: Alias capture to support searching for an object-of-interest is disclosed. A method includes capturing, using a camera with a defined field of view, video image frames that include a moving object-of-interest. The method also includes tracking the object-of-interest over a period of time starting when the object-of-interest enters the field of view and ending when the object-of-interest exits the field of view. The method also includes detecting, at a point in time in-between the start and end of the period of time of the tracking, a threshold exceeding change in an appearance of the object-of-interest. The method also includes creating, before the end of the period of time of the tracking, a new object profile for the object-of-interest in response to the detecting of the threshold exceeding change.

Abstract: A method, system and computer program product for interactively identifying same individuals or objects present in video recordings is disclosed. When a thumbnail in a set of thumbnails is selected, new information is obtained. The new information may be that an individual or object is present in the portion of the video recording associated with the thumbnail. A search can be carried out for the individual or object based on the new information. The search generates new match likelihoods for each of displayed thumbnails within a user interface page. The displayed thumbnails are re-ordered based on the new match likelihoods.

Abstract: Methods, systems, and techniques for sharing layers between convolutional neural networks (CNNs). A data processing system may include a first and a second CNN that share a first group of layers. The first CNN may include the first group of layers in series with a second group of layers and be configured such that data for the first CNN is input to the first group of layers. The second CNN may include the first group of layers in series with a third group of layers and be configured such that data for the second CNN is also input to the first group of layers.

Abstract: A method, system and computer program product for interactively identifying same individuals or objects present in video recordings is disclosed. When a thumbnail in a set of thumbnails is selected, new information is obtained. The new information may be that an individual or object is present in the portion of the video recording associated with the thumbnail. A search can be carried out for the individual or object based on the new information. The search generates new match likelihoods for each of displayed thumbnails within a user interface page. The displayed thumbnails are re-ordered based on the new match likelihoods.

Abstract: Methods, systems, and techniques for monitoring an object-of-interest within a region involve receiving at least data from two sources monitoring a region and correlating that data to determine that an object-of-interest depicted or represented in data from one of the sources is the same object-of-interest depicted or represented in data from the other source. Metadata identifying that the object-of-interest from the two sources is the same object-of-interest is then stored for later use in, for example, object tracking.

Abstract: There are described methods and systems for facilitating identification of an object-of-interest. A face similarity score and a body similarity score of a query image relative to a gallery image are determined. A fused similarity score of the query image relative to the gallery image is determined by applying a relationship between the face similarity score, the body similarity score, and the fused similarity score. The fused similarity score is indicative of whether or not the object-of-interest and the potential object-of-interest are the same object-of-interest. For example, a machine learning process is used to fuse the face similarity score and the body similarity into the fused similarity score. The process is repeated for multiple gallery images. The gallery images may then be ranked according to their respective fused similarity scores.

Abstract: A method, system and computer program product for interactively identifying same individuals or objects present in video recordings is disclosed. When a thumbnail in a set of thumbnails is selected, new information is obtained. The new information may be that an individual or object is present in the portion of the video recording associated with the thumbnail. A search can be carried out for the individual or object based on the new information. The search generates new match likelihoods for each of displayed thumbnails within a user interface page. The displayed thumbnails are re-ordered based on the new match likelihoods.