Implementations of the subject matter described herein provide a solution for thin object detection based on computer vision technology. In the solution, a plurality of images containing at least one thin object to be detected are obtained. A plurality of edges are extracted from the plurality of images, and respective depths of the plurality of edges are determined. In addition, the at least one thin object contained in the plurality of images is identified based on the respective depths of the plurality of edges, the identified at least one thin object being represented by at least one of the plurality of edges. The at least one thin object is an object with a significantly small ratio of cross-sectional area to length. It is usually difficult to detect such thin object with a conventional detection solution, but the implementations of the present disclosure effectively solve this problem.

An unauthorized visitor system collects an image of a person detected in a room of a patient. The system identifies reference points on the person's face, for example, points along the cheeks, jowls, and/or brow. The system may compare the reference points to reference points of images associated with registered visitors. The system then determines, based on the comparison, if the person is a registered visitor. One or more designated recipients may be alerted if the person is not a registered visitor or if the person breaches a patient identification zone established around a particular patient. The system may also register the person in a database of visitors.

Joint coding of depth map video and texture video is provided, where a motion vector for a texture video is predicted from a respective motion vector of a depth map video or vice versa. For scalable video coding, depth map video is coded as a base layer and texture video is coded as an enhancement layer(s). Inter-layer motion prediction predicts motion in texture video from motion in depth map video. With more than one view in a bitstream (for multiview coding), depth map videos are considered monochromatic camera views and are predicted from each other. If joint multiview video model coding tools are allowed, inter-view motion skip is used to predict motion vectors of texture images from depth map images. Furthermore, scalable multiview coding is utilized, where inter-view prediction is applied between views in the same dependency layer, and inter-layer (motion) prediction is applied between layers in the same view.

An apparatus for controlling a plurality of imaging sensor nodes producing 3D structure of a scene is provided. The apparatus receives (500) location data from the sensor nodes, the location data indicating the locations of the moving objects, compares (502) the location data received from different sensor nodes at the same time instants with each other and determines (504) which detections of different sensor nodes relate to same moving objects. The apparatus further maps (600) the location data received from different sensor nodes to a common coordinate system and determines (602) the relationships of the fields of view of the sensor nodes with each other and the location data mapped to the common coordinate system.

In accordance with some embodiments, systems, methods and media for determining object motion in three dimensions using light field image data are provided. In some embodiments, a system for three dimensional motion estimation is provided, comprising: an image sensor; optics that create many images of a scene; and a hardware processor configured to: cause the image sensor to capture a first plurality of images; generate a first light field; cause the image sensor to capture a second plurality of images at a second time; generate a second light field; calculate light field gradients using the first light field and second light field; and calculate, for each point in the scene, three dimensional motion using the light field gradients by applying a constraint to the motion in the scene.

A stroke detection system analyzes images of a person's face over time to detect asymmetric changes in the position of certain reference points that are consistent with sagging or drooping that may be symptomatic of a stroke or TIA. On detecting possible symptoms of a stroke or TIA, the system may alert caregivers or others, and log the event in a database. Identifying stroke symptoms automatically may enable more rapid intervention, and identifying TIA symptoms may enable diagnostic and preventative care to reduce the risk of a future stroke.

An image splicing method includes obtaining a first overlapping image and a second overlapping image from a first image and a second image to-be-spliced. The method also includes determining a motion vector from each pixel in the first overlapping image to a corresponding pixel in the second overlapping image, to obtain an optical flow vector matrix; according to the optical flow vector matrix, remapping the first overlapping image to obtain a first remapping image, and remapping the second overlapping image to obtain a second remapping image; and merging the first remapping image and the second remapping image, to obtain a merged image of the first overlapping image and the second overlapping image, and determining a spliced image of the first image and the second image according to the merged image.

Techniques are provided for calculating temporally coherent disparity values for pixels in a sequence of image frames. An example method may include calculating initial spatial disparity costs between a pixel of a first image frame from a reference camera and pixels from an image frame from a secondary camera. The method may also include estimating a motion vector for the pixel of the first reference camera image frame to a corresponding pixel from a second reference camera image frame. The method may further include calculating a confidence value for the estimated motion vector based on a measure of similarity between the colors of the pixels of the first and second image frames from the reference camera. The method may further include calculating temporally coherent disparity costs based on the initial spatial disparity costs weighted by the confidence value and selecting a disparity value based on those costs.

Systems and/or methods for, for a given pixel (or sub-pixel location) in an image acquired by the camera, finding which projector pixel (or more particularly, which projector column) primarily projected the light that was reflected from the object being scanned back to this camera position (e.g. what projector coordinates or projector column coordinate correspond(s) to these camera coordinates).

Described herein are sensing methods, sensor systems, and non-transitory, computer-readable, storage media having programs for long-duration, continuous monitoring of flying objects during the day or the night and regardless of weather conditions. The methods and systems are computationally efficient and can provide compact, three-dimensional representations of motion from the observed object. A 3D track of the flying object can be generated from a point-matched pair of stereo composite motion track images and not directly from the videos, wherein each composite motion track image is based on a composite of a plurality of video frames composited in part according to video frame numbers.