70 results about "Inverse projection" patented technology

An image processing apparatus projecting 3D image data to 2D planar image data includes: an accumulation unit that accumulates the 3D image data having position coordinates and pixel values; an acquisition unit that acquires a display parameter, including a zoom parameter for changing image size, for the 2D image data to be created; a creation unit that creates the 2D image data from the 3D image data with the display parameter by determining a half view angle of the 3D image data and performing inverse projection transformation on the 2D image data while changing, in accordance with a change in the half view angle caused by the change in image size specified by the zoom parameter, an inverse projection transformation method applied to position coordinates used to create the 2D image data; and a display unit that displays the created 2D image data as a 2D planar image.

A fisheye lens is required to be adopted by the vehicle panoramic display system to obtain a 180-degree wide visual field, however, the visual field is deformed severely. The invention discloses a fisheye image correction method of the vehicle panoramic display system based on a spherical projection model and an inverse transformation model. The method comprises the steps as follows: (1), confirming an area A which is required to be displayed in an aerial view, and establishing a world coordinate system to position the display area; (2), obtaining a coordinate B of the area required to be displayed in a spherical longitude and latitude mapping coordinate system according to camera mounting parameters; (3), confirming the position C of the coordinate B in an original image collected by the fisheye lens according to the fisheye lens spherical projection model; (4), establishing a coordinate transformation relation from C to A through inverse projection transformation; and (5), performing interpolation arithmetic for non-integer points through a bilinear interpolation method to obtain a complete aerial view in a certain direction of the vehicle. The algorithm provided by the invention is simple in implementation and strong in universality and instantaneity.

The invention relates to a method for extraction and deviation warning of a lane line. The method is characterized in that the method comprises the following steps that: an image preprocessing step is executed; more specifically, a camera is calibrated to obtain an inverse projection transmission picture, a single-line DLD feature image and a double-line DLDLD feature image are extracted, operators are extracted according to the two features and binarization processing is carried out on the operators, and clustering expansion processing is carried out on white pixels in the binarization images from bottom to up; continuous linear piecewise function fitting is carried out on the feature points in a predefined grid and coordinates of starting points and end points as well as corresponding gradients of all the piecewise functions are recorded; dashed and solid line determination is carried out on the feature points; and a multi-segment tracking step, a multi-segment extension step, an inspection step and a lane line determination step are carried out to determine a lane, thereby making a deviation warning decision. Advantages of the method are as follows: the lane line detection algorithm is suitable for various complicated urban roads; and because less computation is required and thus operation can be carried out under the circumstance of low-end hardware, the method is simple, secure and reliable.

The invention discloses a method and device for determining a lane line. The method comprises the following steps: performing inverse projection transformation on a photographed image; extracting rough extraction image data including a lane line to be determined from image data after inverse projection transformation; inputting the rough extraction image data into a first convolutional neural network model to recognize, and obtaining a first recognition result of the lane line to be determined; performing inverse projection transformation on the rough extraction image data, inputting the image data of the lane line to be determined after inverse transformation into a second convolutional neural network model to recognize, and obtaining a second recognition result of the lane line to be determined; and determining the lane line to be determined which satisfies a first preset condition as a true lane line according to the first recognition result and the second recognition result. The embodiment of the invention can accurately locate lane lines in the photographed images under various shooting scenes.

The invention discloses an image fusion method of a double / multiple base inverse synthetic aperture radar (ISAR), comprising the following steps: equally dividing target echo signals received by all bases and acquiring two range-Doppler (RD) images; extracting more than three more stable specially obvious points from the target RD images acquired from all the bases, and estimating visual angle differences of any two RD images; combining the visual angle differences and obtaining the estimation of rotation speed of the target and the estimation of the visual angle differences among the double / multiple bases; after obtaining the estimation of the rotation speed of the target, estimating an equivalent rotating center of the target; after knowing the target motion information, obtaining scaling RD images of data of all the bases; also obtaining a polar coordinate imaging result or convolution / inverse projection imaging result of the data of all the bases; and obtaining an imaging result of double / multiple base data level fusion.

The invention provides a visual measuring method for the space target three-dimensional pose of two-phase iterative solutions which are absolute oriented question solution and focus depth estimation adopted in three-dimensional pose calculation. The invention adopts a visual measurement mode which is arranging characteristic cursor point on a target to be measured, and comprises the following steps: camera calibrating, target imaging, image processing, characteristic point extracting and matching and three-dimensional pose calculating. The three-dimensional pose calculation is an iterative process, wherein, the iterative process is based on projecting line inversion and comprises two phases which are absolute oriented question solution and focus depth estimation. The three-dimensional pose calculation comprises the following steps: calculating the relative pose of a space target by adopting an absolute oriented solving analytical algorithm in the absolute oriented solution phase; and in the focus depth estimation phase, reconstructing the object space coordinates of each characteristic point by utilizing the relative pose calculated in the former phase, and updating the focus depth of each characteristic point by utilizing the projection thereof on a reverse projecting line. The three-dimensional pose calculation which adopts a sloving mode of two iteratedly carried out analyzing algorithms has the advantages that the accuracy is high; the convergence is fast; the calculation amount is small; and the applicable range is wide.

Disclosed is a method for accurately and efficiently detecting a modality of input data, including the steps of projecting the input data into a plurality of projection data using each of a plurality of transformation matrix groups U1·(Σ12U2T), generating a plurality of inverse projection data by performing inverse projection of the transformation matrix groups on the plurality of generated projection data, calculating a correlation between the input data and the generated inverse projection data with respect to each transformation matrix group U1·(Σ12U2T), and identifying a modality represented by a transformation matrix group having a highest calculated correlation as the modality of the input data.

The invention provides a high-spectral data noise reduction method and system based on spatial correlation. The high-spectral data noise reduction method comprises the steps of performing projection transformation, namely solving an average image of all images formed by all wave bands in high-spectral data, calculating a covariance matrix of the high-spectral data, performing eigen value decomposition to obtain a transformational matrix and an eigen value matrix, utilizing the transformational matrix to conduct linear projection on the high-spectral data so as to obtain three-dimensional data in a transformation domain; selecting a noise reduction threshold value according to the eigen value matrix; performing noise reduction, namely retaining the three-dimensional data smaller than the noise reduction threshold value in the transformation domain and performing noise reduction processing on the three-dimensional data out of a threshold value range in the transformation domain; performing inverse projection transformation, namely utilizing an inverse matrix of the transformational matrix to conduct linear projection on the three-dimensional data subjected to noise reduction in the transformation domain and obtaining the high-spectral data subjected to noise reduction through reconstruction. In the noise reduction process, the correlation of a spatial domain is fully protected, the noise magnitude of the high-spectral data is well reduced, and the signal to noise ratio of the high-spectral data is improved.

The invention discloses a gene characteristic extraction method based on manifold learning and a closed loop deep convolutional dual-network model. The method comprises the following steps: 1, coarse extraction of a cancer-related gene characteristic based on manifold learning; 2, fine extraction of a gene characteristic based on a closed loop deep convolutional dual-network structure, wherein the process of the fine extraction is as follows: by adopting a dual-network structure consisting of a forward convolutional neural network and a backward convolutional neural network, performing deep abstraction on gene expression data through the characteristic extraction capacity of the convolutional neural networks, and finally projecting key characteristics, wherein the backward convolutional neural network realizes inverse projection of the key characteristics. The invention provides the gene characteristic extraction method based on manifold learning and the closed loop deep convolutional dual-network model, which can retain the gene characteristic to the maximum extent and realize fast dimension reduction.

A traffic situation is predicted based on the correlation in the traffic situation between road sections. A base vector generation unit generates the base vectors constituting a feature space representing the correlation between a plurality of links by making a principal component analysis for the necessary time in the past recorded in a necessary time database. A projection point trajectory generation unit records a projection point trajectory of projecting the necessary time in the past recorded in the necessary time database to the feature space in a projection point database. A feature space projection unit projects the necessary time at present to the feature space, and a neighboring projection point retrieval unit retrieves a past projection point in the neighborhood of the concerned projection point from the projection point database, and a projection point trajectory trace unit traces the trajectory of past projection points starting from the retrieved neighboring projection point for a prediction target time width, and an inverse projection unit inversely projects the end point of the concerned trajectory to calculate the predicted value of the necessary time.

The invention relates to a zebra crossing detection method, and belongs to the technical field of road environment perception in unmanned driving. The method comprises the following steps: S1, detecting a zebra crossing in real time by adopting a camera fixedly mounted at the top of a vehicle to obtain an image; S2, preprocessing the image; S3, performing inverse projection transformation on the zebra crossing candidate area, converting the image into a top view, and performing zebra crossing feature extraction and detection on the basis of the top view; S4, calculating the distance between the peak values, and keeping the peak values with the same distance; S5, determining a zebra crossing candidate area; s6, after determining the zebra crossing candidate area, extracting zebra crossing feature points; and S7, determining a zebra crossing area. According to the invention, through histogram peak value and LSD straight line detection and in combination with black and white mutation point characteristics of the zebra crossing, zebra crossing detection of a complex road section can be provided for an intelligent vehicle, so that the market prospect and the application value are wide.

The invention discloses a vehicle type recognition method based on inverse projection three views. The vehicle type recognition and vehicle modeling method based on the inverse projection three views is not limited by hardware environment, can extract out three-dimension structure information and true size data of a vehicle to be tested and can accurately determine the vehicle type of the vehicle to be tested. Furthermore, a vehicle three-dimension wireframe model can be constructed according to the extracted vehicle structure matched with the inverse projection three views. In addition, the method cannot be affected by vehicle shadow, can accurately and stably recognize the vehicle types according to the three-dimension structure of the vehicle to be tested and is wide in application prospect.

A traffic situation is predicted based on the correlation in the traffic situation between road sections. A base vector generation unit generates the base vectors constituting a feature space representing the correlation between a plurality of links by making a principal component analysis for the necessary time in the past recorded in a necessary time database. A projection point trajectory generation unit records a projection point trajectory of projecting the necessary time in the past recorded in the necessary time database to the feature space in a projection point database. A feature space projection unit projects the necessary time at present to the feature space, and a neighboring projection point retrieval unit retrieves a past projection point in the neighborhood of the concerned projection point from the projection point database, and a projection point trajectory trace unit traces the trajectory of past projection points starting from the retrieved neighboring projection point for a prediction target time width, and an inverse projection unit inversely projects the end point of the concerned trajectory to calculate the predicted value of the necessary time.

The present disclosure concerns a method for reconstructing a point cloud representing a 3D object from an inverse projected point cloud (IPPC) obtained by inverse-projecting at least one depth image of an original point cloud, said at least one depth image being obtained by projecting the points of the original point cloud onto at least one surface, said method comprising the steps of; detecting (900) at least one missing part (MPtfj in the inverse-projected point cloud (IPPC), and completing (910) said at least one missing part (MPq) based on points in the neighborhood of said at least one missing part.

The invention discloses a Treelet-based Bayer type CFA (Color Filter Array) image denoising method which mainly solves the problem that a large amount of residual noise occurs easily when the existing denoising method is directly applied to a CFA image. The method comprises the following steps: exacting high frequency information from an input noisy CFA image; performing blocking processing on the high frequency image, gradually exacting training data from all image blocks, and projecting training data to a Treelet basis matrix; obtaining denoised high frequency image blocks by contracting projection coefficients and performing inverse projection; and splicing all the denoised high frequency image blocks, and adding low frequency information input to the CFA image, so as to obtain a denoised CFA image. The method can directly perform adaptive denoising on the CFA image, simultaneously can reduce the residual noise in the denoised image in the condition that image detail is properly kept, and can be applied to the denoising processing on the CFA images captured through a single CCD (Charge Couple Device) or CMOS (Complementary Metal Oxide Semiconductor) sensor camera.

A method and system generates a reconstructed gray scale image of pixels from a binary image of pixels by reverse propagating an error, in a diffused manner, to the pixels of the binary image to create pixels representing a reconstructed gray scale image. The reverse propagation of the error determines if a current pixel has a first predetermined value; sets an error value to a first error value if the current pixel has the first predetermined value; sets the error value to a second error value if the current pixel does not have the first predetermined value; and propagates a diffused error value to a pixel neighboring the current pixel to create an error diffused pixel, the diffused error value being a product of the set error value and a weighting coefficient associated with the pixel neighboring the current pixel. A Gaussian filter may be utilized to filter the pixels representing a gray scale image to create a filtered reconstructed gray scale image of pixels.

The invention aims to provide the light irradiation therapeutic apparatus capable of accurately positioning the disease area of a patient and irradiating the most proper light to the disease area in order to reach the best treatment effect. The apparatus provided by the invention comprises one or more cameras and an image collection apparatus, a light irradiating system and a computing and processing unit and is characterized in that the image of the disease area of the patient is collected by one or more cameras; three-dimensional match of the disease area is built according to the space coordinate relation of multiple cameras or a projection characteristic image of the light irradiation system; a three-dimensional structure of the disease area of the patient and the position relation relative to the cameras are obtained; a light irradiation dose table needed by the disease area is generated according to a light irradiation therapeutic scheme constituted by a doctor aiming at the disease area; an inverse projection therapeutic scheme is built by the computing and processing unit according to a dose table and the coordinate relation between the light irradiation system and the cameras, so as to control the light irradiation system to realize the irradiation with the predetermined dose in the disease area; and the actual light irradiation amount of the light irradiation system can be synchronously monitored by the one or more cameras.

The invention discloses a method for correcting an intersecting line of an unstructured Cartesian grid. The method comprises the following steps of: describing the shape of an object to generate a background grid; setting grid generation parameters; generating an n-ary tree structure according to the background grid and the set parameters; deleting units within the shape of the object and units intersected with the shape of the object to generate a saw-toothed inner surface; fairing the saw-toothed inner surface to obtain a faired inner surface; performing intersecting line correction and projection on the faired inner surface; and calculating a grid flux, and outputting grid information. The problem of object surface distortion possibly caused by the steps of fairing, projection and the like when the grid is generated in computational fluid mechanics is solved, and a finally generated computational grid object surface can be ensured to be approximate to an original object shape in an intersecting line inverse projection and correction way.

A CPU generates an image by performing inverse projection transformation for an input image by using an inverse matrix of a matrix which enlarges a size of the input image. When a position A on the input image is transformed into a position B on the generated image by the inverse projection transformation, the CPU obtains a ratio of a distance between pixels around the position B to a distance between pixels around the position A. The CPU calculates filter coefficients for a low-pass filter using a product of a Nyquist frequency of the input image and the ratio as a cutoff frequency. The CPU obtains a sum of results obtained by multiplying pixel values of pixels in a pixel array including the position A in the input image by the filter coefficients, as a pixel value at the position A in an image having undergone filter processing.

For the purpose of providing a backprojection method and an X-ray CT apparatus by which backprojection processing can be simplified and sped up, instead of obtaining backprojection pixel data D2(x, y) directly from projection data D0(view, ch), axially projected data D1(view, pt) are obtained from projection data D0(view, ch) by projecting them onto a straight line, and then backprojection pixel data D2 are obtained from the axially projected data D1.

A CPU generates an image by performing inverse projection transformation for an input image by using an inverse matrix of a matrix which enlarges a size of the input image. When a position A on the input image is transformed into a position B on the generated image by the inverse projection transformation, the CPU obtains a ratio of a distance between pixels around the position B to a distance between pixels around the position A. The CPU calculates filter coefficients for a low-pass filter using a product of a Nyquist frequency of the input image and the ratio as a cutoff frequency. The CPU obtains a sum of results obtained by multiplying pixel values of pixels in a pixel array including the position A in the input image by the filter coefficients, as a pixel value at the position A in an image having undergone filter processing.

Methods and apparatus of processing 360-degree virtual reality images are disclosed. According to one method, a system that maps a motion vector in a 2D frame into a 3D space and then applies rotationof sphere to adjust motion vector in the 3D space. The rotated motion vector is then mapped back to the 2D frame for processing 360-degree virtual reality images. According to another method, a system that derives motion vector from displacement of viewpoint is disclosed. According to yet another method, a system that applies scaling to adjust motion vector is disclosed. The motion vector scalingcan be performed in the 3D space by projecting a motion vector from a 2D frame to a 3D sphere. After motion vector scaling, the scaled motion vector is projected back to the 2D frame. Alternatively,the forward projection, motion vector scaling and inverse projection can be combined into a single function.

The invention discloses a multiband ultra-wideband imaging processing method for near-field measurement of an MIMO radar. The method comprises the main steps that a filtering-inverse projection (MIMO-FBP) algorithm suitable for near-field measurement of the MIMO (multiple-input-multiple-output) radar is used to carry out image reconstruction on multiband imaging data of the MIMO radar, and channelequalization, directional diagram correction, propagation attenuation compensation and spherical wavefront correction are completed automatically in each band to obtain low-resolution complex imagesin the different bands; 2D fast Fourier transform (2D-FFT) is carried out to transform the complex images in the different bands to corresponding data domains, re-sampling is carried out along a sector ring, and sector ring domain data corresponding to the images in the different bands is obtained; and the multiband sector ring domain imaging data is fused in a proper manner to obtain full waveband MIMO ultra-wideband imaging data; and a traditional FBP imaging algorithm is used to carry out image reconstruction on the fused full waveband data. Thus, a full waveband high-resolution image is obtained.

A method for encoding a 3D point cloud comprises: segmenting the point cloud into first patches (2); - determining inverse projection parameters of the patches (4); projecting the first patches to planes using the inverse projection parameters (10); encoding the planes (12); determining sparse points in the point cloud not belonging to any patch (HC, 18); and encoding the sparse points using an octree-decomposition (20).

The invention provides a method for automatically adjusting back focal length (BFL) of an unthreaded lens based on an inverse projection mode. The problem that a lens seat is mounted through a thread and is focused is solved. According to the method, the BFL (back focal length) of the lens is adjusted to a micron level in the inverse projection mode before a module is assembled, so that the assembled module is not required to be focused; and therefore, the aim of avoiding focusing is fulfilled, and the production efficiency is improved. A camera shooting module can be subjected to automatic unthreaded focusing, the method has the advantages of high accuracy, high efficiency and the like, the influence of subjective consciousness of an operator on the product quality is avoided during manual focusing, the consistency and stability of the product quality are improved, the burden of the operator and the labor cost of an enterprise are reduced, and compared with the traditional scheme, the method has the advantage that the overall dimension of the product can be small.

The invention discloses a fire-fighting area passability analysis system based on stereoscopic vision. The firefighting area passability analysis system comprises an input module, a depth estimation module and a passability analysis module, the depth estimation module comprises a twin network layer with shared weight, a pyramid stereo matching network, a projection transformation layer and a three-dimensional inverse projection layer; the twin network layer and the pyramid stereo matching network receive the binocular image transmitted by the input module, and pixel-level features for stereo matching and advanced features for target detection are extracted; the projection transformation layer is used for constructing a plane scanning body; the three-dimensional inverse projection layer isused for converting the plane scanning body into a 3D geometry through a derivable image torsion operation, and constructing three-dimensional geometrical characteristics of a world coordinate system;and a passability analysis module which is constructed based on a 3D convolutional network and is used for performing passability analysis on the three-dimensional geometry. According to the invention, the scene depth can be jointly generated, the passability analysis is carried out on the 4 * 4 * 10 m< 3 > space of the fire-fighting area, and the smoothness of a fire-fighting life passage is ensured along with the patrol of the intelligent miniature fire-fighting vehicle-mounted binocular vision system.

The invention relates to an avoidance judgment method and apparatus for emergency vehicles at an intersection, and a medium. The method comprises that: a road image is obtained in advance; the image is processed to obtain the position of a solid line, and the position is calibrated; a vehicle coming direction road video is shot and stored; video frames are extracted, vehicle detection is carried out on the road image through a target detection technology based on deep learning, and meanwhile, emergency vehicles needing to be avoided are identified; according to the relative positions in the same lane, whether the target vehicle possibly avoids the emergency vehicle is judged; if yes, the image is transformed into a top view by using inverse projection transformation; whether line pressing occurs or not is judged; if yes, illegal line pressing can be judged, and then avoidance judgment is carried out; and if yes, license plate recognition is carried out, and information is uploaded to execute violation record clearing. According to the invention, the efficiency of violation appeal work of avoiding the emergency vehicle can be improved, a guarantee is provided for a driver of the avoiding vehicle, and the initiative of avoiding the emergency vehicle can be improved.

Disclosed is a method for accurately and efficiently detecting a modality of input data, including the steps of projecting the input data into a plurality of projection data using each of a plurality of transformation matrix groups U1·(Σ12U2T), generating a plurality of inverse projection data by performing inverse projection of the transformation matrix groups on the plurality of generated projection data, calculating a correlation between the input data and the generated inverse projection data with respect to each transformation matrix group U1·(Σ12U2T), and identifying a modality represented by a transformation matrix group having a highest calculated correlation as the modality of the input data.

The invention relates to the field of projection devices, and particularly relates to a slide for oblique projection, a making method thereof and a projection lens. The making method for the slide for oblique projection comprises the following steps: S1, according to a projection view field, the mounting position of the projection lens and the view field angle of the projection lens obtained based on requirements, a projection model is built; S2, a Sensor surface vertical to the optical axis of the projection lens in the projection model is made, and an actual image which needs to be obtained is put in the projection view field range; S3, an inverse projection method is adopted to project the actual image in the projection view field range on the Sensor surface; and S4, after the image on the Sensor surface is zoomed and adjusted, the image is transferred to the slide to make the slide. Fidelity of a projection graph at an oblique angle can be realized, and being clear and sharp is realized.