Abstract: Systems and methods for map construction using a video sequence captured on a camera of a vehicle in an environment, comprising: receiving a video sequence from the camera, the video sequence including a plurality of image frames capturing a scene of the environment of the vehicle; using a neural camera model to predict a depth map and a ray surface for the plurality of image frames in the received video sequence; and constructing a map of the scene of the environment based on image data captured in the plurality of frames and depth information in the predicted depth maps.

Abstract: Disclosed herein are systems, methods, and non-transitory computer readable mediums directed to identifying persons monitored on video frames of a video feed, determining if the number of people on a monitored video frame is above a prescribed group threshold, and if so, calculating distances between each respective person in the monitored video frame. The calculated distances between each respective person are then compared to a prescribed distance threshold and alerts are generated in response to any calculated distance not satisfying the prescribed threshold. Identification of persons in a monitored video frame may be performed by, for example, facial recognition software. Calculation of distance between persons in a monitored video frame may be performed, for example, by using bounding boxes, machine learning, or other classification techniques.

Abstract: A system and method for making categorized music tracks available to end user applications. The tracks may be categorized based on computer-derived rhythm, texture and pitch (RTP) scores for tracks derived from high-level acoustic attributes, which is based on low level data extracted from the tracks. RTP scores are stored in a universal database common to all of the music publishers so that the same track, once RTP scored, does not need to be re-RTP scored by other music publishers. End user applications access an API server to import collections of tracks published by publishers, to create playlists and initiate music streaming. Each end user application is sponsored by a single music publisher so that only tracks capable of being streamed by the music publisher are available to the sponsored end user application.

Abstract: Techniques for training multiple resolution deep neural networks (DNNs) for vehicle autonomous driving comprise obtaining a training dataset for training a plurality of DNNs for an autonomous driving feature of the vehicle, sub-sampling the training dataset to obtain a plurality of training datasets comprising the training dataset and one or more sub-sampled datasets each having a different resolution than a remainder of the plurality of training datasets, training the plurality of DNNs using the plurality of training datasets, respectively, determining a plurality of outputs for the autonomous driving feature using the plurality of trained DNNs and the input data, receiving input data for the autonomous driving feature captured by a sensor device, and determining a best output for the autonomous driving feature using the plurality of outputs.

Abstract: A structural vibration monitoring method based on computer vision and motion compensation provided in the present disclosure adopts a dual-camera system for self-motion compensation. The dual-camera system consists of a primary camera and a secondary camera rigidly connected to each other. The primary camera directly measures a structure displacement. This method inevitably includes an error generated due to motion of the primary camera. Meanwhile, the secondary camera measures displacements of translation and rotation, so as to estimate a measurement error caused by the motion of the primary camera. Then, with the displacement directly measured by the main camera minus the measurement error, a corrected structure displacement is obtained, thereby truthfully and accurately monitoring vibrations of a bridge structure.

Abstract: Embodiments are disclosed for using an apparatus to perform pupil identification. An example method includes toggling circuitry configured to determine, based on a measured ambient infrared light level, whether to perform pupil identification using one of either the dark pupil identification process or the bright pupil identification process; and pupil identification circuitry configured to perform pupil identification on a user using the bright identification process upon the determination that the measured ambient infrared light level falls within a first range and to perform dark pupil identification process upon the determination that the measured ambient infrared light level falls within a second range. Corresponding apparatuses and non-transitory computer readable storage media are also provided.

Abstract: A method of performing an integrity check on a plurality of video surveillance cameras includes retrieving a recent video frame from each of the plurality of video surveillance cameras, determining a integrity score for each of the recent video frames and determining which of the integrity scores failed to meet a predetermined criteria. When at least one of the integrity scores fails to meet the predetermined criteria, an integrity check report is created that includes an camera identifier along with the recent video frame for each of the plurality of video surveillance cameras that had a recent video frame with an integrity score that failed to meet the predetermined criteria. The integrity check report is displayed to a user for manual verification.

Abstract: Calculating, for each frame of a plurality of frames, a corresponding quality value; calculating, for each frame of the plurality of frames, based on one or more visual attributes of a frame, a weight for the corresponding quality value of the frame; calculating an aggregate quality value for the plurality of frames based on the weight and the corresponding quality value for each frame of the plurality of frames; and providing an assessment of the plurality of frames based on the aggregate quality value for the plurality of frames.

Abstract: A lamp and a method for detecting a user's sitting posture are provided. The lamp is adapted to detect the user's sitting posture. The lamp comprises a lamp bracket, an image capturing unit and a processing unit. The image capturing unit is disposed on the lamp bracket and configured to capture a target image towards a desktop. The processing unit is coupled to the image capturing unit, wherein in response to that a head feature exists in the target image, the processing unit is configured to determine a reference information according to the head feature and to determine whether the reference information meets a poor sitting posture condition.

Abstract: This invention provides a system and method for finding patterns in images that incorporates neural net classifiers. A pattern finding tool is coupled with a classifier that can be run before or after the tool to have labeled pattern results with sub-pixel accuracy. In the case of a pattern finding tool that can detect multiple templates, its performance is improved when a neural net classifier informs the pattern finding tool to work only on a subset of the originally trained templates. Similarly, in the case of a pattern finding tool that initially detects a pattern, a neural network classifier can then determine whether it has found the correct pattern. The neural network can also reconstruct/clean-up an imaged shape, and/or to eliminate pixels less relevant to the shape of interest, therefore reducing the search time, as well significantly increasing the chance of lock on the correct shapes.

Abstract: The subject matter of this specification can be implemented in, among other things, methods, systems, computer-readable storage medium. A method includes a processing device receiving image data having one or more image frames indicative of a state of a meal preparation area. The method further includes the processing device determining, based on the image data, one of a meal preparation item or a meal preparation action associated with the state of the meal preparation area. The method further includes the processing device determining, based on the image data, a performance metric associated with the meal preparation item or the meal preparation item or the meal preparation action. The method further includes the processing device causing a notification corresponding to the performance metric to be displayed on a graphical user interface (GUI).

Abstract: Physical object boundary detection techniques and systems are described. In one example, an augmented reality module generates three dimensional point cloud data. This data describes depths at respective points within a physical environment that includes the physical object. A physical object boundary detection module is then employed to filter the point cloud data by removing points that correspond to a ground plane. The module then performs a nearest neighbor search to locate a subset of the points within the filtered point cloud data that correspond to the physical object. Based on this subset, the module projects the subset of points onto the ground plane to generate a two-dimensional boundary. The two-dimensional boundary is then extruded based on a height determined from a point having a maximum distance from the ground plane from the filtered cloud point data.

Abstract: An information processing apparatus (2000) detects an abnormal region (30) from a predetermined range (16) of a video frame (14). The information processing apparatus (2000) determines whether a predetermined condition is satisfied in a case where the abnormal region (30) is detected from the predetermined range (16) of a certain video frame (14) and the abnormal region (30) is not detected from the predetermined range (16) of a predetermined video frame (14) generated later than the video frame (14). In a case where the predetermined condition is not satisfied, the information processing apparatus (2000) performs a predetermined notification.

Abstract: Systems and methods for scalable comparisons between two pixel maps are provided. In an embodiment, an agricultural intelligence computer system generates pixel maps from non-image data by transforming a plurality of values and location values into pixel values and pixel locations. The non-image data may include data relating to a particular agricultural field, such as nutrient content in the soil, pH values, soil moisture, elevation, temperature, and/or measured crop yields. The agricultural intelligence computer system converts each pixel map into a vector of values. The agricultural intelligence computer system also generates a matrix of metric coefficients where each value in the matrix of metric coefficients is computed using a spatial distance between to pixel locations in one of the pixel maps. Using the vectors of values and the matrix of metric coefficients, the agricultural intelligence computer system generates a difference metric identifying a difference between the two pixel maps.

Abstract: A method includes generating a first warped image based on a pose and a depth estimated from a current image and a previous image in a sequence of images captured by a camera of the agent. The method also includes estimating a motion of dynamic object between the previous image and the target image. The method further includes generating a second warped image from the first warped image based on the estimated motion. The method still further includes controlling an action of an agent based on the second warped image.

Abstract: A system and method is disclosed for classifying natural language sentences by employing external knowledge to assist in constructing a knowledge base of sentences with a target meaning. The disclosed system and method provide a general sentence classification framework applicable for a knowledge-oriented domain (e.g., domain-specific knowledge). The system and method may be implemented in an intelligent automotive aftermarket assistance tool to assist with the identification of sentences describing specific problems and solutions for car repairs. In addition to the domain adaptability, the system and method is language-independent and could be applicable to any natural written language.

Abstract: Disclosed are a video stitching method and a video stitching device. The video stitching method is applicable for stitching a first video and a second video, and includes: performing feature extraction, feature matching and screening on a first target frame of the first video and a second target frame of the second video, so as to obtain a first feature point pair set; performing forward tracking on the first target frame and the second target frame, so as to obtain a second feature point pair set; performing backward tracking on the first target frame and the second target frame, so as to obtain a third feature point pair set; and calculating a geometric transformation relationship between the first target frame and the second target frame according to a union of the first feature point pair set, the second feature point pair set and the third feature point pair set.

Abstract: Apparatus and methods for applying motion blur to overcapture content. In one embodiment, the motion blur is applied by selecting a number of frames of the captured image content for application of motion blur; selecting a plurality of pixel locations within the number of frames of the captured image content for the application of motion blur; applying motion blur to the captured image content in accordance with the selected number of frames and the selected plurality of pixel locations; and outputting the captured image content with the applied motion blur. In some implementations, motion blur is applied via implementation of a virtualized neutral density filter. Computerized devices and computer-readable apparatus for the application of motion blur are also disclosed.

Abstract: Systems, devices, and methods for training a neural network and performing image interest point detection and description using the neural network. The neural network may include an interest point detector subnetwork and a descriptor subnetwork. An optical device may include at least one camera for capturing a first image and a second image. A first set of interest points and a first descriptor may be calculated using the neural network based on the first image, and a second set of interest points and a second descriptor may be calculated using the neural network based on the second image. A homography between the first image and the second image may be determined based on the first and second sets of interest points and the first and second descriptors. The optical device may adjust virtual image light being projected onto an eyepiece based on the homography.

Abstract: A deep neural network(s) (DNN) may be used to detect objects from sensor data of a three dimensional (3D) environment. For example, a multi-view perception DNN may include multiple constituent DNNs or stages chained together that sequentially process different views of the 3D environment. An example DNN may include a first stage that performs class segmentation in a first view (e.g., perspective view) and a second stage that performs class segmentation and/or regresses instance geometry in a second view (e.g., top-down). The DNN outputs may be processed to generate 2D and/or 3D bounding boxes and class labels for detected objects in the 3D environment. As such, the techniques described herein may be used to detect and classify animate objects and/or parts of an environment, and these detections and classifications may be provided to an autonomous vehicle drive stack to enable safe planning and control of the autonomous vehicle.