2671 results about "Rgb image" patented technology

An apparatus and method for image data compression performs a modified zero-tree coding on a range of image bit plane values from the largest to a defined smaller value, and a vector quantizer codes the remaining values and lossless coding is performed on the results of the two coding steps. The defined smaller value can be adjusted iteratively to meet a preselected compressed image size criterion or to meet a predefined level of image quality, as determined by any suitable metric. If the image to be compressed is in RGB color space, the apparatus converts the RGB image to a less redundant color space before commencing further processing.

The invention discloses a positioning method and system based on visual inertial navigation information fusion. The method comprises steps as follows: acquired sensor information is preprocessed, wherein the sensor information comprises an RGB image and depth image information of a depth vision sensor and IMU (inertial measurement unit) data; external parameters of a system which the depth visionsensor and an IMU belong to are acquired; the pre-processed sensor information and external parameters are processed with an IMU pre-integration model and a depth camera model, and pose information isacquired; the pose information is corrected on the basis of a loop detection mode, and the corrected globally uniform pose information is acquired. The method has good robustness in the positioning process and the positioning accuracy is improved.

The subject disclosure presents systems and computer-implemented methods for automatic immune cell detection that is of assistance in clinical immune profile studies. The automatic immune cell detection method involves retrieving a plurality of image channels from a multi-channel image such as an RGB image or biologically meaningful unmixed image. A cell detector is trained to identify the immune cells by a convolutional neural network in one or multiple image channels. Further, the automatic immune cell detection algorithm involves utilizing a non-maximum suppression algorithm to obtain the immune cell coordinates from a probability map of immune cell presence possibility generated from the convolutional neural network classifier.

The invention discloses an image semantic segmentation method based on deep learning. The method comprises four parts of data set processing, deep semantic segmentation network construction, deep semantic segmentation network training and parameter learning, and semantic segmentation on a test image. The RGB image and the gray level image of the input image are used as the input of the network model, the edge information of the gray level image is fully utilized, and the richness degree of input characteristics is effectively increased; a convolutional neural network and a bidirectional threshold recursion unit are combined, and more context dependency relationships and global feature information are captured on the basis of learning image local features; coordinate information is added tothe feature map through the first coordinate channel module and the second coordinate channel module, the coordinate features of the model are enriched, the generalization ability of the model is improved, and a semantic segmentation result with high resolution and accurate boundary is generated.

The invention discloses an autonomous location and navigation method and an autonomous location and navigation system of a robot. The method comprises the following steps: acquiring the RGB image of the robot in the current position; carrying out characteristic point matching on the RGB image of the current position and an indoor three-dimensional point cloud map, and determining the current position of the robot in the indoor 3D point cloud map; carrying out characteristic point matching on a pre-stored RGB image of a destination and the indoor three-dimensional point cloud map, and determining the position of the destination in the indoor 3D point cloud map; and searching an optimal path from the current position to the destination position in the indoor 3D point cloud map, and driving the robot to run according to the optimal path. The method and the system have the advantages of completion of autonomous location and navigation through using a visual sensor, simple device structure, low cost, simplicity in operation, and high path planning real-time property, and can be used in the fields of unmanned driving and indoor location and navigation.

The invention discloses a full convolution neural network (FCN)-based monocular image depth estimation method. The method comprises the steps of acquiring training image data; inputting the training image data into a full convolution neural network (FCN), and sequentially outputting through pooling layers to obtain a characteristic image; subjecting each characteristic image outputted by a last pooling layer sequentially to amplification treatment to obtain a new characteristic image the same with the dimension of a characteristic image outputted by a previous pooling layer, and fusing the twocharacteristic images; sequentially fusing the outputted characteristic image of each pooling layer from back to front so as to obtain a final prediction depth image; training the parameters of the full convolution neural network (FCN) by utilizing a random gradient descent method (SGD) during training; acquiring an RGB image required for depth prediction, and inputting the RGB image into the well trained full convolution neural network (FCN) so as to obtain a corresponding prediction depth image. According to the method, the problem that the resolution of an output image is low in the convolution process can be solved. By adopting the form of the full convolution neural network, a full-connection layer is removed. The number of parameters in the network is effectively reduced.

The invention discloses a bidirectional long short-term memory unit-based behavior identification method for a video. The method comprises the steps of (1) inputting a video sequence and extracting an RGB (Red, Green and Blue) frame sequence and an optical flow image from the video sequence; (2) respectively training a deep convolutional network of an RGB image and a deep convolutional network of the optical flow image; (3) extracting multilayer characteristics of the network, wherein characteristics of a third convolutional layer, a fifth convolutional layer and a seventh fully connected layer are at least extracted, and the characteristics of the convolutional layers are pooled; (4) training a recurrent neural network constructed by use of a bidirectional long short-term memory unit to obtain a probability matrix of each frame of the video; and (5) averaging the probability matrixes, finally fusing the probability matrixes of an optical flow frame and an RGB frame, taking a category with a maximum probability as a final classification result, and thus realizing behavior identification. According to the method, the conventional artificial characteristics are replaced with multi-layer depth learning characteristics, the depth characteristics of different layers represent different pieces of information, and the combination of multi-layer characteristics can improve the accuracy rate of classification; and the time information is captured by use of the bidirectional long short-term memory, many pieces of time domain structural information are obtained and a behavior identification effect is improved.

The present invention discloses a system and method for non-holographic virtual teleportation of one or more remote objects to a designated three-dimensional space around a user in realtime. This is achieved by using a plurality of participating teleportation terminals, located at geographically diverse locations, that capture the RGB and depth data of their respective environments, extract RGB images of target objects from their corresponding environment, and transmit the alpha channeled object images via Internet for integration into a single composite scene in which layers of computer graphics are added in the foreground and in the background. The invention thus creates a virtual 3D space around a user in which almost anything imaginable can be virtually teleported for user interaction. This invention has application in intuitive computing, perceptual computing, entertainment, gaming, virtual meetings, conferencing, gesture based interfaces, advertising, ecommerce, social networking, virtual training, education, so on and so forth.

An embodiment of the present application discloses a method, a device, equipment and a storage medium for detecting human face in vivo. The method comprises the following steps of: acquiring an RGB image and a depth image of a target object; calling a first neural network to process the RGB image to obtain a first processing result; invoking a second neural network to process the depth image to obtain a second processing result; according to the first processing result and the second processing result, determining whether or not the target object is a human face living body. An embodiment of the present application obtains an RGB image and a depth image of a target object, combining the two images to detect the face of the target object in vivo, the face texture clues and 3D face structureclues being combined to detect the face in vivo, which improves the accuracy and can effectively intercept the face paper, reproduce the face on the high-definition screen, synthesize the face video,face mask, 3D prosthetic model and other attacks. Moreover, it is simpler and more efficient to capture RGB images and depth images without user interaction.

A depth estimation method for a monocular image based on a multi-scale CNN and a continuous CRF is disclosed in this invention. A CRF module is adopted to calculate a unary potential energy according to the output depth map of a DCNN, and the pairwise sparse potential energy according to input RGB images. MAP (maximum a posteriori estimation) algorithm is used to infer the optimized depth map at last. The present invention integrates optimization theories of the multi-scale CNN with that of the continuous CRF. High accuracy and a clear contour are both achieved in the estimated depth map; the depth estimated by the present invention has a high resolution and detailed contour information can be kept for all objects in the scene, which provides better visual effects.

An RGB-Z sensor is implementable on a single IC chip. A beam splitter such as a hot mirror receives and separates incoming first and second spectral band optical energy from a target object into preferably RGB image components and preferably NIR Z components. The RGB image and Z components are detected by respective RGB and NIR pixel detector array regions, which output respective image data and Z data. The pixel size and array resolutions of these regions need not be equal, and both array regions may be formed on a common IC chip. A display using the image data can be augmented with Z data to help recognize a target object. The resultant structure combines optical efficiency of beam splitting with the simplicity of a single IC chip implementation. A method of using the single chip red, green, blue, distance (RGB-Z) sensor is also disclosed.

The invention relates to a stacked workpiece posture recognizing and picking method based on an RGBD camera. The method includes the following steps that 1), parameters in the RGBD camera are calibrated; 2), training is carried out according to a pre-obtained 3D model of a to-be-captured workpiece, and a 2D model for matching is generated; 3), an RGB image and a depth image of the to-be-recognizedworkpiece are obtained through the RGBD camera, and contour information of the to-be-captured workpiece is obtained; 4), two-dimensional position information of the to-be-captured workpiece in an image pixel coordinate system and a six-freedom-degree pose under a camera coordinate system are obtained; 5), a six-freedom-degree pose of the to-be-captured workpiece under a robot coordinate system isobtained; and 6), a six-axis robot is controlled to pick the to-be-captured workpiece. Compared with the prior art, the RGBD camera with low cost is utilized, in combination with RGB and depth information, the postures of various types of workpieces stacked disorderly are recognized and captured, and the method is high in precision, low in cost and high in adaptability, and can meet the requirements of industrial production.

The invention discloses an indoor robot SLAM method. The method comprises the following steps: obtaining image data through a camera, wherein the image data comprises RGB images and depth images; using a corner detection algorithm, an LK characteristic point tracing algorithm, and a RANSAC algorithm to process the RGB images and depth images to adjust the position and angle of the camera so as to obtain the RGB image data information under a robot operation system; converting the depth images in a world coordinate system to a ground coordinate, traversing the depth images that are projected on the ground coordinate, setting the gray value of the grid where a barrier stays as a first characteristic value, carrying out traversing to obtain a 2D barrier grid map; through a single line laser radar scanning mode, searching the 2D barrier grid map, when the grid with a grey value equal to the first characteristic value is found, feeding back the distance between the grid and the camera to obtain the distance between a robot and the barrier, obtaining the depth images under a robot operation system, and obtaining an environment 2D map.

The invention discloses a method and a device for correcting color based on an RGBIR (red, green and blue, infra red) image sensor. The method comprises the following steps of entering an RGBIR image; interpolating the image, and calculating the channel values of an RGB (red, green and blue) channel and an IR (infra red) channel on each pixel position; calculating the channel statistical values of an RGB channel and an IR channel in a particular area; according to the statistical values, looking up the table to obtain the correction coefficients of the current scene; configuring the correction coefficients into a correction matrix, and separating the RGB channel and the IR channel; carrying out the color correction on the separated RGB channel, to obtain the corrected RGB image. A pre-correcting unit eliminates the effect of invisible light on the RGB channel of the original image, and the output utilizes a back-correcting unit to eliminate the effect of wavelength change of visible light on the RGB channel. The method not only can correct the color of the RBGIR image sensor, but also can divide the color correction into two relatively independent parts, and can be well compatible with the color correction method in the existing image processing module.

Systems, devices, and methods that integrate eye tracking capability into scanning laser projector (“SLP”)-based wearable heads-up displays are described. At least one narrow waveband laser diode is used in an SLP to define one or more portion(s) of a visible image. At least one corresponding narrow waveband photodetector is aligned to detect reflections of the portion(s) of the image from features of the eye. A holographic optical element (“HOE”) may be used to combine the image and environmental light into the user's “field of view.” Three narrow waveband photodetectors each responsive to a respective one of three narrow wavebands output by the RGB laser diodes of an RGB SLP are aligned to detect reflections of a projected RGB image from features of the eye.

The invention discloses a dynamic gesture recognition method and system based on a deep neural network. The dynamic gesture recognition method comprises the steps of collecting dynamic gesture video clips with different gesture meanings to generate a training sample data set, wherein the sample data includes RGB images and depth information; designing a dynamic gesture recognition network model based on the deep neural network, and training the model by using the training samples; and performing dynamic gesture testing and recognition by using the trained dynamic gesture recognition model. Thedynamic gesture recognition network model is composed of a feature extraction network, a front and back frame association network and a classification recognition network, wherein the front and backframe association network is used for performing front and back time frame association mapping on feature vectors obtained through the feature extraction network of the samples of each gesture meaningand merging the feature vectors into a fusion feature vector of the gesture meaning. According to the invention, a bidirectional LSTM model is introduced into the network model to understand the correlation between continuous gesture postures, thereby greatly improving the recognition rate of dynamic gestures.

The invention relates to a nondestructive detection method for comprehensive character living bodies of plant seedlings, which comprises the following steps of: performing high-spectrum imaging on the plant seedlings, extracting RGB images and a spectrum of a specific area from a high-spectrum data cube, and acquiring morphological parameter, component content distribution and disease and insect pest information of the seedlings by image processing and spectrum analysis; and establishing a growth prediction model of the plant seedlings by adopting a character-level information fusion method and a fuzzy comprehensive judgment method. The method can realize comprehensive judgment on characters of appearance, nutritional component and disease and insect pest information of the seedlings, and overcome the error caused by un-uniformity during component measurement.

An embodiment of the invention provides a method and device for detecting lane lines on a road surface. The method includes the steps of: collecting an RGB image of a road surface and a point cloud intensity map of the road surface; performing lane line detection on the RGB image and the point cloud intensity map respectively, thereby obtaining image three-dimensional lane lines in the RGB image and point cloud three-dimensional lane lines in the point cloud intensity map; and based on a preset lane line fusing strategy, fusing the image three-dimensional lane lines and the point cloud three-dimensional lane lines, thereby obtaining three-dimensional lane lines of the road surface, and outputting the three-dimensional lane lines. By fusion of a lane line detection result of the RGB image and a lane line detection result of the point cloud intensity map, shielding of the road surface lane lines by other vehicles and wear of the lane lines on the road surface and road surface direction arrows can be effectively resisted in most cases, integrity and precision of geometric outlines of the obtained three-dimensional lane lines of the road surface are improved, and in addition, manpower cost is saved.

The invention provides a man-machine cooperation oriented real-time posture detection method for hand-held objects, belonging to the technical field of a human-machine cooperation interaction system and the position and posture sensing of an industrial robot to a hand-held working object. The depth images of each part of the object to be detected are captured by a 3D stereoscopic camera, and the local point clouds are aligned and merged into a complete three-dimensional point cloud model of the object. Real-time RGB color images and depth images containing 3D point cloud information of the scene are obtained. The RGB image is segmented automatically to get the pixels representing the object in the image. The corresponding point cloud in the depth image is fused with the pixels to get the RGB-D image with color information of the object in the scene. Using ICP algorithm, the RGB-D image is matched with the complete 3D point cloud image of the object to obtain the position and posture ofthe hand-held object in the scene. The method overcomes the problem of obtaining the exact posture of the hand-held object at the current time, and can be used in a variety of scenes.

The present invention presents a framework for separating specular and diffuse reflection components in images and videos. Each pixel of the an M-channel input image illuminated by N light sources is linearly transformed into a new color space having (M-N) channels. For an RGB image with one light source, the new color space has two color channels (U,V) that are free of specularities and a third channel (S) that contains both specular and diffuse components. When used with multiple light sources, the transformation may be used to produce a specular invariant image. A diffuse RGB image can be obtained by applying a non-linear partial differential equation to an RGB image to iteratively erode the specular component at each pixel. An optional third dimension of time may be added for processing video images. After the specular and diffuse components are separated, dichromatic editing may be used to independently process the diffuse and the specular components to add or suppress visual effects. The (U,V) channels of images can be used as input to 3-D shape estimation algorithms including shape-from-shading, photometric stereo, binocular and multinocular stereopsis, and structure-from-motion.

Under the premise of no extraneous high-resolution image library, a super-resolution sparse representation method for acquiring a high-resolution image is provided. The method comprises the steps of: (1) carrying out space conversion on a given low-resolution color image to obtain the YCbCr space image of the color image, and reconstructing the constituents of Cb and Cr by using an interpolation method; (2) constructing a database used for training, namely, a high-resolution image block Xh and a low-resolution image block Xl, and combining the two image blocks into a database X; (3) generating a dictionary D from the database X by using a sparse coding method, decomposing the dictionary D into a high-resolution image dictionary Dh and a low-resolution image dictionary Dl; (4) solving a sparse coefficient by using the Dl and characteristic images corresponding to an image of upsampling the low-resolution image by 2 times; (5) solving an image of upsampling the original low-resolution image by 3 times through the sparse coefficient and the Dh; and (6) combining Y, Cb, and Cr to obtain a YCbCr image, and converting the YCbCr image into an RGB image and storing the RGB image to obtain the final super-resolution representation image.