Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for retroactive event detection. The methods, systems, and apparatus include actions of obtaining an image captured by a camera at a current time, determining that the image depicts a change in a region from a previous image captured by the camera at a previous time, determining, based on determining that the image depicts the change in the region, whether the change depicted in the image is of a known object type, determining, based on the determination that the change depicted in the image is of a known object type, whether the change does not correspond to a previously detected event, and determining, based on the determination that the change does not correspond to a previously detected event, whether the images captured by the camera between the current time and the previous time depict an event.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training an event detector. The methods, systems, and apparatus include actions of identifying a portion of a first interframe difference image that represents motion of an OI, determining that a second interframe difference image represents motion by a non-OI, combining the portion of the first interframe difference image and the second interframe difference image as a third interframe difference image labeled as motion of both an OI and a non-OI, and training an event detector with the third interframe difference image.

Abstract: Methods and systems, including computer programs encoded on a storage medium, are described for implementing item monitoring using a doorbell camera. A system generates an input video stream that has image frames corresponding to detection of activity at a property. Timing information is generated for the video stream and includes a timestamp for each image frame of the stream. Using the timing information, the system processes a pre-event image frame that precedes detection of the activity and a post-event image frame that coincides with detection of the activity. An image score is computed with respect to placement of a candidate item at the property in response to processing the pre-event and post-event image frames. The image score is used to determine that a first item was delivered to the property or that a second item was removed after being delivered to the property.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for using feature descriptors to track objects depicted in images. One of the methods includes receiving hue, saturation, value data for an image and data that indicates an object detected in the image, generating a feature descriptor that includes hue data and saturation data, determining, for each of two or more tracked objects that each have a historical feature descriptor that includes historical hue data and historical saturation data, a distance between (i) the respective historical feature descriptor and (ii) the feature descriptor, associating the feature descriptor for the object with a tracked object from the two or more tracked objects, and tracking the tracked object in one or more images from a video sequence using the feature descriptor and the historical feature descriptor.

Abstract: Methods, systems, and apparatus for removing precipitation from video are disclosed. A method includes generating, from a first set of images of a scene from a camera, a segmented background image model of the scene; obtaining a second set of images from the camera; identifying, in an image of the second set of images, a plurality of edges, determining that a first edge of the plurality of edges satisfies criteria for representing precipitation based at least in part on determining that the first edge (i) does not correspond to the background image model of the scene and (ii) extends into two or more contiguous segments of the scene; in response, classifying each of the contiguous segments as a precipitation segment; generating pixel data for each of the precipitation segments; and applying the pixel data to each precipitation segment in the image.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training an event detector. The methods, systems, and apparatus include actions of obtaining frames of a video, determining whether an object of interest is detected within the frames, determining whether motion is detected within the frames, determining whether the frames correspond to motion by an object of interest, generating a training set that includes labeled inter-frame differences based on whether the frames correspond to motion by an object of interest, and training an event detector using the training set.

Abstract: Methods, and systems including computer programs encoded on a computer storage medium, for training a detection model for surveillance devices using semi-supervised learning. In one aspect, the methods include receiving imaging data collected by a camera of a scene within a field of view of the camera. Annotated training data is generated from the imaging data and one or more detection models are trained using the annotated training data. Based on a set of performance parameters, an optimized detection model is selected of the one or more detection models, and the optimized detection model is provided to the camera.

Abstract: Methods, systems, and apparatus for camera detection of human activity with co-occurrence are disclosed. A method includes detecting a person in an image captured by a camera; in response to detecting the person in the image, determining optical flow in portions of a first set of images; determining that particular portions of the first set of images satisfy optical flow criteria; in response to determining that the particular portions of the first set of images satisfy optical flow criteria, classifying the particular portions of the first set of images as indicative of human activity; receiving a second set of images captured by the camera after the first set of images; and determining that the second set of images likely shows human activity based on analyzing portions of the second set of images that correspond to the particular portions of the first set of images classified as indicative of human activity.

Abstract: Methods, systems, and apparatus for ground plane filtering of video events are disclosed. A method includes obtaining a first set of images of a scene from a camera; determining a ground plane from the first set of images of the scene; obtaining a second set of images of the scene after the first set of images of the scene is obtained; determining that movement shown by a group of pixels in the second set of images of the scene satisfies motion criteria; determining that the ground plane corresponds with at least a portion of the group of pixels; and in response to determining that movement shown by the group of pixels in the second set of images of the scene satisfies motion criteria, and that the ground plane corresponds with at least a portion of the group of pixels, classifying the group of pixels as showing ground plane based motion.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training an event detector. The methods, systems, and apparatus include actions of obtaining frames of a video, determining whether an object of interest is detected within the frames, determining whether motion is detected within the frames, determining whether the frames correspond to motion by an object of interest, generating a training set that includes labeled inter-frame differences based on whether the frames correspond to motion by an object of interest, and training an event detector using the training set.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for classifying actions of an object. The methods, systems, and apparatus include actions of: obtaining frames of video including an object of interest; determining a type of action of the object in each of the frames of video; determining a group of frames from the frames of video based on the type of action; determining an aggregated background subtraction (ABS) image based on adjacent frames of the group of frames; generating a training set that includes labeled ABS images including the ABS image; and training an action classifier using the training set.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for preserving privacy in surveillance. The methods, systems, and apparatus include actions of obtaining images of a scene captured by a camera, identifying an object in the images through object recognition, determining that the object that is identified in the images is of a particular type that has a privacy restriction, and in response to determining that the object in the images is of the particular type that has the privacy restriction, obfuscating an appearance of the object in the images.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for preserving privacy in surveillance. The methods, systems, and apparatus include actions of obtaining images of a scene captured by a camera, identifying an object in the images through object recognition, determining that the object that is identified in the images is of a particular type that has a privacy restriction, and in response to determining that the object in the images is of the particular type that has the privacy restriction, obfuscating an appearance of the object in the images.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for preserving privacy in surveillance. The methods, systems, and apparatus include actions of obtaining images of a scene captured by a camera, identifying an object in the images through object recognition, determining that the object that is identified in the images is of a particular type that has a privacy restriction, and in response to determining that the object in the images is of the particular type that has the privacy restriction, obfuscating an appearance of the object in the images.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for virtual inductance loop technology. In one aspect, a method includes calibrating, using calibration parameters, cameras directed towards a scene, obtaining, by the cameras, images corresponding to the scene, identifying reference structures in the scene, and determining, based on the reference structures and the images, locations in the scene for generating virtual inductance loop lines in the scene. The method also includes generating the virtual inductance loop lines to be imposed on the images, comparing the virtual inductance loop lines to determine one or more offsets, and determining, based on the offsets, characteristics of the scene.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for preserving privacy in surveillance. The methods, systems, and apparatus include actions of obtaining images of a scene captured by a camera, identifying an object in the images through object recognition, determining that the object that is identified in the images is of a particular type that has a privacy restriction, and in response to determining that the object in the images is of the particular type that has the privacy restriction, obfuscating an appearance of the object in the images.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for virtual inductance loop technology. In one aspect, a method includes calibrating, using calibration parameters, cameras directed towards a scene, obtaining, by the cameras, images corresponding to the scene, identifying reference structures in the scene, and determining, based on the reference structures and the images, locations in the scene for generating virtual inductance loop lines in the scene. The method also includes generating the virtual inductance loop lines to be imposed on the images, comparing the virtual inductance loop lines to determine one or more offsets, and determining, based on the offsets, characteristics of the scene.

Abstract: Methods and systems for facial recognition are provided. The method includes determining a three-dimensional (3D) model of a face of an individual based on different images of the individual. The method also includes extracting two-dimensional (2D) patches from the 3D model. Further, the method includes generating a plurality of signatures of the face using different combinations of the 2D patches, wherein the plurality of signatures correspond to respective views of the 3D model from different angles.

Abstract: Methods and systems for facial recognition are provided. The method includes determining a three-dimensional (3D) model of a face of an individual based on different images of the individual. The method also includes extracting two-dimensional (2D) patches from the 3D model. Further, the method includes generating a plurality of signatures of the face using different combinations of the 2D patches, wherein the plurality of signatures correspond to respective views of the 3D model from different angles.

Abstract: Embodiments relate to systems, devices, and computer-implemented methods for predicting social media trends by receiving multiple sets of social media data from a social media service, wherein each set of social media data includes multiple entries and each entry is associated with a user identifier. For each set of social media data: labels can be extracted; a social media data graph can be generated with nodes representing labels and user identifiers and edges representing a co-occurrence of labels or a co-occurrence of a label and a user identifier; and the social media data graph can be analyzed to determine a graph metric score for nodes corresponding to a label. The graph metric scores of a node across multiple sets of social media data can be used to predict that the label corresponding to the node will be significant to trending, e.g., will begin trending.