91 results about "Perspective distortion" patented technology

Perspective distortion is estimated in a digital document image by detecting perspective pencils in two directions, one being parallel to text lines, and the other being parallel to the boundaries of formatted text columns. The pencils are detected by analyzing directional statistical characteristics of the image. To detect a pencil, a first statistical line transform is applied to transform the image into line space, and a second statistical score transform is applied to transform the image into pencil space. A dominant peak in pencil space identifies the perspective pencil. In addition, a computationally efficient line summing technique is used for effecting sums of pixels along inclined target lines (or curves) through an image. The technique includes pre-generating partial sums, and summing along step segments of a target line using the partial sums.

Arrangements are detailed to aid a user in positioning a camera-equipped device (e.g., a smartphone) at a particular viewing position relative to an object being imaged. Other embodiments discern the device's viewing position, and then undertake image or other processing that is dependent on such viewpoint. A great number of other features and arrangements are also detailed.

Perspective distortion is estimated in a digital document image by detecting perspective pencils in two directions, one being parallel to text lines, and the other being parallel to the boundaries of formatted text columns. The pencils are detected by analyzing directional statistical characteristics of the image. To detect a pencil, a first statistical line transform is applied to transform the image into line space, and a second statistical score transform is applied to transform the image into pencil space. A dominant peak in pencil space identifies the perspective pencil. In addition, a computationally efficient line summing technique is used for effecting sums of pixels along inclined target lines (or curves) through an image. The technique includes pre-generating partial sums, and summing along step segments of a target line using the partial sums.

Object recognition by point-of-sale camera systems is aided by first removing perspective distortion. Yet pose of the object—relative to the system—depends on actions of the operator, and is usually unknown. Multiple trial counter-distortions to remove perspective distortion can be attempted, but the number of such trials is limited by the frame rate of the camera system—which limits the available processing interval. One embodiment of the present technology examines historical image data to determine counter-distortions that statistically yield best object recognition results. Similarly, the system can analyze historical data to learn what sub-parts of captured imagery most likely enable object recognition. A set-cover strategy is desirably used. In some arrangements, the system identifies different counter-distortions, and image sub-parts, that work best with different clerk- and customer-operators of the system, and processes captured imagery accordingly. A great variety of other features and arrangements are also detailed.

An electronic imaging system for capturing an image of a scene includes an optical system for producing an optical image of the scene, an imaging sensor having a surface in optical communication with the optical system, and a plurality of imaging elements distributed on the surface of the imaging sensor according to a distribution representable by a nonlinear function in which the relative density of the distributed imaging elements is greater toward the center of the sensor. Such a distribution provides physical coordinates for the imaging elements corresponding to a projection of the scene onto a non-planar surface, thereby compensating for perspective distortion of the scene onto the non-planar surface and alleviating the need to perform geometric warping of the images after they have been captured.

The invention provides an image processing method and an associated device for correcting a fisheye image and reducing perspective distortion. The fisheye image processing method is used for correcting the fisheye image correction and reducing the perspective distortion and can save the space of a temporary storage and reduce the perspective distortion. The method comprises the following steps: sequentially calculating a fisheye original coordinate value corresponding to each pixel of a plurality of pixels in the fisheye image according to a coordinate conversion function between a fisheye corrected image and a fisheye original image; and calculating the pixel value of each pixel of the fisheye corrected image by interpolation in the fisheye original image according to the fisheye original coordinate value, wherein the coordinate conversion function contains a fisheye coordinate correcting conversion function, a perspective distortion reducing coordinate conversion function and an image cutting zoom-in and zoom-out coordinate conversion function.

The invention discloses a water level gauge positioning and water level measuring method based on image processing, wherein the water level gauge positioning method comprises the following steps: calculating the longitudinal inclination angle of a mark rod main body of a pre-extracted water level gauge image; obtaining a water level gauge main body image after inclination correction by adopting arotation correction technology according to the longitudinal inclination angle of the mark rod main body; correcting perspective distortion by using a calibration point technology for the main image of the water level gauge after inclination correction; using a projection method to carry out adjacent pixel repair on a missing value part in the distortion-corrected image; carrying out accurate target clipping on the repaired image according to the coordinate equation of the left and right dividing lines; the starting position of the water level gauge and the position of the boundary line with the water surface are identified and the main body position of the water level gauge is cut according to the starting position and the position of the boundary line with the water surface. According tothe water level gauge positioning and water level measuring method based on image processing, the distortion correction is carried out on the water level gauge image after the inclination correction,the distortion problem caused by the non-parallel lateral boundary of the target caused by the viewpoint selection angle of the image is solved, and the precision of target extraction is improved.

The invention discloses a metal plate strip product label information identification method based on computer vision. the position of a product label area is obtained through segmentation of a lightweight network; the coordinate information of the product label is obtained through an image processing means; correction to enabline see-through transformer is realized, the VGG16 is used for identifying the rotating text; character rotation small-angle registration is carried out by using a variance method; the text position detection precision and the text recognition precision are effectively improved; YOLOv3 and ENet are adopted, so that text correction and position acquisition are faster and more accurate; the loss of the computer and the requirement on the performance of the computer areeffectively reduced; the detection of the text with the uncertain length is realized by utilizing the characteristics of the LSTM in the CRNN; The detection performance is effectively improved, the good recognition performance is achieved in natural scenes such as non-uniform illumination, complex backgrounds, multi-language mixing, text complex formats, product label picture rotation, affine distortion and perspective distortion, and convenience is provided for inputting of label information of metal plate strips.

The invention discloses a positioning correction part in a QR code detection process, and a positioning and correction algorithm on the condition that one of three image seeking patterns of the QR code misses. The algorithm binarizes the image, then finds out the other two images besides the damaged image and calculates the version number of QR code. For version 1 QR codes, morphological transformation and edge extraction are carried out, and then the vertices are detected by line detection, and then corrected. For QR codes with version number 2 and above, the correction pattern of the lower right corner in the standard pattern of the QR code is searched, and the correction transformation is carried out by using the center points of the image seeking patterns and the center point of the correction pattern. The invention can accurately position and correct the QR codes on the condition that one of the image seeking patterns of the QR code misses and certain perspective distortion exists.

An image processing apparatus and an image processing method are provided. The image processing method is implemented by the apparatus, which receives and stores captured image data. Addresses for the captured image data are generated and stored in a look up table in memory along with color signal data that is stored in an additional data area. Output image data is generated by interpolating the address information in the lookup table to determine coordinate information for the output image. The output image coordinate information allows for drawing an output image with corrected image distortion, corrected perspective distortion, altered viewpoint from captured image, mirror-image conversion, or electronic zooming of the captured image. Color signal data from the additional data area is then used to draw an overlay on the output image in color.

The invention discloses a tool setting method of a mechanical arm feeding type laser etching system, belongs to the field of non-traditional machining, and relates to a tool setting method of a mechanical arm feeding type laser etching system based on a machine vision technology. The laser etching system is optimized and improved; a vision sensor and an active projection indication laser are added; and an interactive information obtaining function between heterogeneous space non-cooperative electromechanical equipment in the system is added. The problems of non-front-view perspective distortion and vision sensor imaging distortion introduced by an inclined vision measurement configuration are considered; and an inverse perspective transformation geometric correction and camera distortion compensation technology is adopted to correct image information. The precise alignment between a laser etching focusing plane and a part initial characteristic point to be machined in a working space is realized by combining multi-frame sequence image dynamic transformation and multi-axis driving feeding device space shape and position characteristic information. The method solves the precise positioning tool setting problem in the laser etching machining initial phase, and improves the machining precision of target workpieces and the finished product quality.

Arrangements are detailed to aid a user in positioning a camera-equipped device (e.g., a smartphone) at a particular viewing position relative to an object being imaged. Other embodiments discern the device's viewing position, and then undertake image or other processing that is dependent on such viewpoint. A great number of other features and arrangements are also detailed.

The invention discloses a self-adaptive receptive field crowd density estimation method based on cavity convolution, which belongs to the field of computer vision, and comprises the following steps: segmenting an original data set image and a crowd density map to obtain image blocks and crowd density map blocks; constructing and training an adaptive receptive field population density estimation network, wherein the model comprises a cavity convolution module and a classification module, the classification module is used for classifying the segmented image blocks, the cavity convolution moduleadaptively selects a cavity convolution sub-network corresponding to the receptive field according to the image block category output by the classification module, and performs feature extraction on the segmented image blocks to obtain a crowd density map; and inputting the picture to be predicted into the trained adaptive receptive field crowd density estimation model to obtain a crowd density estimation result. According to the method, the cavity convolution sub-network corresponding to the receptive field can be adaptively selected for crowd density estimation, and the problem of perspective distortion is solved, so that the accuracy of crowd density estimation is improved.

The invention discloses a curved surface image restoration method, device and equipment and a readable storage medium, and the method comprises the steps: carrying out the preprocessing of a first curved surface two-dimensional code image after the first curved surface two-dimensional code image is obtained, and obtaining a second curved surface two-dimensional code image; carrying out plane perspective correction operation on the second curved surface two-dimensional code image to obtain a third curved surface two-dimensional code image; and performing curved surface restoration operation onthe third curved surface two-dimensional code image to obtain a planar two-dimensional code image corresponding to the first curved surface two-dimensional code image. According to the invention, theshot curved-surface two-dimensional code image caused by perspective distortion is restored into the planar two-dimensional code image through the planar perspective correction operation and the curved-surface restoration operation.

A system, method, and apparatus for generating a normalization of a single two-dimensional image of an unconstrained human face. The system receives the single two-dimensional image of the unconstrained human face, generates an undistorted face based on the unconstrained human face by removing perspective distortion from the unconstrained human face via a perspective undistortion network, generates an evenly lit face based on the undistorted face by normalizing lighting of the undistorted face via a lighting translation network, and generates a frontalized and neutralized expression face based on the evenly lit face via an expression neutralization network.

A fisheye correction with perspective distortion reduction method used for saving memory space and reducing perspective distortion includes a coordinate transformation function, a fisheye distorted image and a fisheye corrected image, the fisheye corrected image includes a plurality of pixels, for each pixel of the plurality of pixels, orderly calculating a coordinate value in the fisheye distorted image, and according to the coordinate value, calculating a pixel value of the fisheye corrected image via an interpolation method, wherein the coordinate transformation function includes a fisheye correction coordinate transformation, a perspective distortion reduction coordinate transformation and an image crop and scale coordinate transformation.

The invention relates to the technical field of image processing and image recognition, in particular to a table correction and recognition method based on combination of image processing and deep learning, which comprises the following steps: step 110, acquiring original image data of a table; step 120, image preprocessing; step 130, positioning a character area; 140, reconstructing table information; by designing and improving the existing form recognition method, the accuracy of form recognition is improved by performing text direction judgment, inclination correction and perspective distortion processing when the form image is recognized, and the problem that the form recognition accuracy is low after the form image is obtained by using equipment in the existing method for recognizingthe form in the image is solved. The table row and column frame line positions are detected by analyzing the optical characteristics of the whole-page digital image so as to detect the format structure of the table, and the method is generally only suitable for the conditions that the input image quality is good, the table positions and formats are fixed, and the table frame lines are obvious, andthe problems of character direction overturning, inclination, perspective distortion and the like exist in the image.

The invention relates to the technical field of crowd counting methods, and discloses a picture-based crowd counting method, which comprises the following steps of S1, inputting an original picture, and extracting picture features through a VGG-16 network; S2, carrying out 2 * 2 average pooling operation on the extracted VGG feature values to obtain blocks with the size of kxk, and then carrying out convolution on the blocks and a convolution layer with the kernel of 1; the convolution operation with the kernel of 1 has the greatest advantage that the dimension of an original feature value isnot changed, and then a result is input into a Normalization Laye (normalization layer); and S3, re-up-sampling to recover to the feature size of the previous VGG, and finally decoding and outputtingthe density map. According to the picture-based crowd counting method, an end-to-end trainable network architecture is adopted, multi-scale transformation is adapted by learning how to configure weights for different features at an independent pixel level, the influence of perspective distortion on a density map can be effectively reduced, and the accuracy of a crowd calculation result is improved.

The invention discloses a lane guide arrow identification method in an intersection monitoring environment, belongs to the field of computer vision, and relates to image processing related knowledge.The method comprises the steps: acquiring a traffic video image sequence based on an intersection monitoring camera; establishing a lane background model by using a background modeling technology; detecting a lane line by utilizing Hough transform on the basis of the established lane background model, taking an end point of the detected lane line as an original coordinate point of image perspective transformation, and finishing perspective transformation in combination with a least square method to obtain a lane top view, so as to correct perspective deformation generated by shooting of a camera; then, respectively projecting the lane top view in the vertical direction and the horizontal direction, and carrying out screening segmentation according to the ratio of the length of a standard guide arrow to the width of the lane to obtain all target area images only containing a single guide arrow; finally, performing similarity matching on the guide arrows obtained through segmentation anda standard guide arrow template one by one, and recognizing the types of the guide arrows of all lanes.

The invention provides a method for detecting and counting dense crowd distribution in a video. Firstly, acquiring a large number of videos containing crowds with different densities to construct a data set; then constructing a deep neural network of multi-scale feature fusion and an attention mechanism, inputting the training set into the network, outputting prediction results of a corresponding crowd density map and an attention map, constructing a loss function model in combination with the real density map and the attention map for training, and generating an optimized network; obtaining a density map of a crowd video image through optimized multi-scale feature fusion and deep neural network prediction of an attention mechanism, furthering performing point clustering on the estimated density map by using a grid-based hierarchical density space clustering method to identify a group, and obtaining the number of people and position information of the group quickly. According to the invention, the problems of perspective distortion, scale change and background noise influence of the camera can be solved, and the counting precision and stability are improved; and meanwhile, the crowd is divided into groups, so that the distribution condition of the crowd can be visually displayed.

The invention discloses a camera imaging error calibration and correction method. A secondary calibration model is introduced to solve the problem that the correction precision is low due to calibration of a least square method commonly used at present, and camera installation errors cannot be corrected. According to the method, distortion and perspective deformation can be corrected at the same time in a high-precision mode, correction coordinates of characteristic points of the calibration plate completely coincide with ideal coordinates, correction parameters of the characteristic points are introduced into non-characteristic points through an interpolation algorithm, and high-precision correction of the whole image is achieved. Moreover, the traditional calibration method needs to process the images of a plurality of calibration plates, and the method only needs to process the image of one calibration plate, thereby simplifying the calibration process, and greatly reducing the timeand complexity of the calibration work.

The invention provides a license plate image enhancement method and device, equipment and a storage medium. The method comprises the following steps: acquiring a first position relationship among vertexes of a license plate to be enhanced and a second position relationship among vertexes of a license plate position in a to-be-fused background image; calculating a perspective transformation matrix between the to-be-enhanced license plate and the to-be-fused background image according to the first position relation and the second position relation; and fusing the to-be-enhanced license plate to the license plate position of the to-be-fused background image according to the transformation matrix to obtain an enhanced license plate image. According to the method, the perspective distortion is effectively reduced in a mode of fusing the license plate and the real background image through the perspective transformation matrix, and the influence of the defects that the real background is difficult to replace by the virtual construction background and the license plate distortion and the virtual construction background distortion are difficult to keep consistent on the enhanced image is avoided.

The invention discloses an intelligent and accurate inspection method and system for a power tower by an unmanned aerial vehicle. The method comprises the steps: designing and achieving the inclined frame detection of a target based on a target detection algorithm, and extracting an inclined frame target object; designing a target three-dimensional detection network, namely fusing a target inclined frame detection task and a depth estimation network, running on embedded equipment in a multi-thread mode, and obtaining position information and depth information of a target; calculating a pose vector group of the target by using the position information and the depth information of the target; and calculating a distance correction amount and a holder angle correction amount of the unmanned aerial vehicle according to the pose vector group of the target, so that the unmanned aerial vehicle performs inspection photographing after correction. The system comprises an inclined frame target detection unit, a position and depth information acquisition unit, a pose vector group acquisition unit and a distance and holder angle correction unit. According to the invention, the problem of perspective distortion of an image during inspection shooting of the unmanned aerial vehicle can be solved, and the quality of the inspection shooting image is improved.

An image processing method includes searching an image area corresponding to a predetermined image from a photographed image obtained by photographing the predetermined image, which is projected on a projection plane; calculating parameters, which are used for correcting perspective distortion of the photographed image, based on the searched image area; and correcting the photographed image and another image, which is photographed after calculating the parameters, based on the calculated parameters.

The invention discloses a flood routing process flooding range measurement method based on deep learning. The method comprises the steps that firstly, a camera is arranged to be used for collecting video data of a whole test river channel; calibrating the camera through a camera checkerboard calibration method to correct the perspective distortion effect; extracting image data at different time points from the video data; constructing a flood test flooding range sample library; carrying out primary labeling on the sample through a Label labeling tool; and finally, adopting an MASK R-CNN imageinstance segmentation algorithm to realize automatic segmentation and recognition of a flooding range, and obtaining the submerging range change of the whole test riverway by splicing the recognitionpictures. The method has the advantages of being low in economic cost, high in intelligent degree, high in efficiency, high in precision, high in applicability and the like, and therefore the method can be used for extracting the obtained flooding range data in the flood routing test.

The invention relates to the field of three-dimensional morphology measurement, and particularly discloses a camera calibration method and device for a 3D microscopic shape measurement system, and the method comprises the steps: generating a phase target, carrying out the snapshot of a camera, carrying out the phase analysis of a snapshot calibration picture, generating a phase diagram, extracting the feature point of the circle center of an integer pixel from the phase diagram, performing equiphase line extraction in a D * D neighborhood of a pixel size such as 35 * 35 by taking the whole image circle center feature point as a center, performing ellipse fitting to obtain a sub-pixel precision circle center feature point, and calculating a camera parameter initial value according to a telecentric camera calibration algorithm according to a pixel coordinate of the sub-pixel precision circle center feature point and a corresponding world coordinate; and obtaining high-precision camera parameters by using a repeated iteration method. The device and method have very high robustness to image blurring and noise, the target is more accurate, and the extraction of the circle center feature point is not influenced by lens distortion and perspective distortion.