1263 results about "Edge extraction" patented technology

A background subtraction apparatus of the present invention includes a key point locator for locating key points on a known object type, a boundary point locator for locating boundary points of the known object that make up the edges of the known object, and an edge processor for processing the edges to provide a clean-edged extraction of the known object from a background image. Preferably, the key point locator includes an alignment detector for detecting alignment of an image of the known object type with a skeleton image. Still more preferably, the skeleton image is an exoskeleton image and the known object type is a human being.

A data model which is designed for being superposed with an image of a real object in an optical object tracking process is determined by the following steps: providing a three-dimensional CAD model (10) for representing the real object, and thereafter there are different synthetic two-dimensional views (31 to 34) of said CAD model (10) generated. Each generated view (31 to 34) is subjected to edge extraction for determining at least one extracted edge (38, 39) in the respective view, with the edges (38, 39) extracted from said respective views (31 to 34) being transformed to a three-dimensional contour model (85, 91) corresponding to said data model to be determined. Permits rapid and efficient generation of a contour model as a data model intended for being superposed with an image of a real object.

A technique for generating high-resolution bitmaps from low-resolution bitmaps. A low-resolution bitmap is magnified to form a magnified image. Edge detection is performed on the magnified image to find high contrast edges. A plurality of image patches of the magnified image are generated. These images patches are analyzed by performing connected components analysis on each of them using the high contrast edges to produce a plurality of foreground and background decisions determining whether a portion of an image patch is a background or a foreground region. Then the contrast of one or more pixels in each of the plurality of image patches is enhanced based on the foreground and background decisions. Finally, the system and method of the invention combines the luminance of the enhanced output pixels with the color values generated by the magnification algorithm. This produces a high-resolution bitmap from the contrast-enhanced pixels.

The invention relates to an on-line detecting method for surface defects of an object and a device for realizing the method. The accuracy for the detection and the distinguishing of the defects is improved through the fusion of grey and depth information, and the method and the device can be applied to the detection of the object with a complicated shape and a complicated surface. A grey image and a depth image of the surface of the object are collected by utilizing the combination of a single colored area array CCD (charge-coupled device) camera and a plurality of light sources with different colors, wherein obtaining of the depth information is achieved through a surface structured light way. The division and the defect edge extraction of the images are carried out through the pixel level fusion of the depth image and the grey image, so that the area where the defects are positioned can be detected more accurately. According to the detected area with the defects, the grey characteristics, the texture characteristics and the two-dimensional geometrical characteristics of the defects are extracted from the grey image; the three-dimensional geometrical characteristics of the defects are extracted from the depth image; further, the fusion of characteristic levels is carried out; and a fused characteristic quantity is used as the input of a classifier to classify the defects, thereby achieving the distinguishing of the defects.

A detecting and error-correcting method capable of rapidly and accurately obtaining an element center and a deflection angle is disclosed. Interested edge area can be rapidly and accurately found out through an automatic edge area extraction method based on combination of binary morphology and image reduction, and subsequent edge extraction image scope is minimized as possible. Pixel level edge is roughly positioned at first, and accordingly a sub-pixel edge is rapidly extracted by one-dimensional curve fitting method. The method overcomes the defect of complex algorithm of a conventional sub-pixel edge extraction algorithm based on two-dimensional images, and the time of extracting accurate sub-pixel edge can be shortened. Weighted least square rectangular edge fitting algorithm based on linear hazen paradigm can detect straight lines on any positions of an image, and effectively minimizes influence of outlier on fitting precision due to uneven edges. The central positions of a plurality of elements and deflect angles can be rapidly and accurately detected at once, and the efficiency of visual detection is improved.

The invention discloses a garment QD code recognition method. The method includes the following steps that firstly, a two-dimension code image attached to a garment raw material is collected; secondly, the image is subjected to graying, fast median filtering and binarization processing; thirdly, the image having been subjected to binarization processing is subjected to edge extraction; fourthly, two side edge regions obtained after edge extraction of a QR code are separated out; fifthly, Hough transform is adopted to detect a side-edge imaginary line so as to obtain the deviation angle of the QR code; sixthly, the image having been subjected to binarization processing is rotated according to the bilinear interpolation method, and initial correction is carried out; seventhly, the QR code is positioned, the sequence of three view finding images is determined, and the QR code is adjusted to have a correct orientation according to the image matrix transposition turning method; eighthly, decoding is carried out according to the national two-dimension code standard. According to the method, due to the combination of projection cutting and the Hough transform method, the deviation angle of the QR code can be quickly and accurately acquired, the collected image is immune from noise pollution and uneven illumination, and the two-dimension code image easy to identify can be obtained through transform.

The invention discloses a driving safety monitoring system based on face orientation and visual focus, which comprises a vision sensor and an intelligent processor. The intelligent processor comprises an image acquisition module, a complexion area detection module, an edge extraction module, an eye detection module, a face orientation analysis module and a driver safety driving judgment module, wherein the complexion area detection module is used for converting an RGB color space into a YCrCb color space and detecting a complexion area; the edge extraction module is used for adopting a Canny operator to carry out edge extraction so as to acquire a head image outline; the eye detection module is used for carrying out Hough conversion for the marginalized face and positioning eyes; the face orientation analysis module is used for determining a mouth area, positioning the mouth, calculating left and right areas of the face respectively by using the eyes and the mouth as reference, and calculating the ratio of the left and right areas of the face; and the driver safety driving judgment module is used for judging an unsafe driving state if the calculated ratio of the left and right areas of the current face is out of a preset interval according to the preset interval of the ratio of the left and right areas of the face, and sending out an alarm command. The driving safety monitoring system has strong capacity of resisting disturbance and high accuracy.

The invention discloses a deep learning-based remote sensing image semantic segmentation method. The method comprises the steps of assigning an RGB value and a gray value to each target species, obtaining an original remote sensing image, selecting a target species to color and gray the target species, imparting a gray value to the target species to obtain a label image, and subjecting the original remote sensing image to data enhancement and edge extraction to obtain an edge-extracted image; training a full convolution neural network by adopting the original remote sensing image and the imagetraining sample of the edge-extracted image to obtain an optimum semantic segmentation network model, and inputting a to-be-tested remote sensing image into the optimum semantic segmentation networkmodel to obtain a semantic segmentation result image; coloring the semantic segmentation result image to obtain a final semantic segmentation result image, and obtaining a species target according toRGB values in the final semantic segmentation result image. According to the method, the semantic segmentation results of remote sensing images are high in accuracy, and the method is wide in applicability.

Provided is a pattern inspecting and measuring device that decreases the influence of noise and the like and increases the reliability of an inspection or measurement result during inspection or measurement using the position of an edge extracted from image data obtained by imaging a pattern as the object of inspection or measurement. For this purpose, in the pattern inspecting and measuring device in which inspection or measurement of an inspection or measurement object pattern is performed using the position of the edge extracted, with the use of an edge extraction parameter, from the image data obtained by imaging the inspection or measurement object pattern, the edge extraction parameter is generated using a reference pattern having a shape as an inspection or measurement reference and the image data.

The invention discloses a camera lens occlusion detecting system and method. The system comprises an image collection module, an image pre-processing module, a background establish module, a high-frequency component extraction module, a generalized image establish module, a suspected occlusion area detection module and an occlusion area determining module, wherein the background establish module is used for acquiring a smooth background image; the high-frequency component extraction module differentiates an initial image and the background image to acquire a high-frequency component distribution graph; the generalized image establish module combines an enhanced image and a pre-processed initial image to establish a generalized image, and performs edge extraction on the generalized image and the generalized image is convolved; the suspected occlusion area detection module compares a convolution result with a preset threshold, and determines that the suspected occlusion area is a candidate occlusion area if the convolution result is less than the preset threshold; and the occlusion area determining module traces the subsequent image pixel value of the suspected occlusion area, and determines that the suspected occlusion area is an occlusion area finally if brightness variations of all the pixels in the area are less than the preset threshold. By using the system and the method, whether the lens is occluded or not can be effectively judged.

A technique for generating high-resolution bitmaps from low-resolution bitmaps. A low-resolution bitmap is magnified to form a magnified image. Edge detection is performed on the magnified image to find high contrast edges. A plurality of image patches of the magnified image are generated. These images patches are analyzed by performing connected components analysis on each of them using the high contrast edges to produce a plurality of foreground and background decisions determining whether a portion of an image patch is a background or a foreground region. Then the contrast of one or more pixels in each of the plurality of image patches is enhanced based on the foreground and background decisions. Finally, the system and method of the invention combines the luminance of the enhanced output pixels with the color values generated by the magnification algorithm. This produces a high-resolution bitmap from the contrast-enhanced pixels.

The invention discloses a method for detecting an illegal driving behavior based on hand gesture tracking. The method comprises the following steps of: 1, reading an image from a monitoring video; 2, pre-processing the read image, and particularly performing gray conversion, image filtration, edge extraction, contour enhancement and the like; 3, positioning a steering wheel; 4, intercepting a corresponding interested area by using the steering wheel as the center; 5, extracting features of the intercepted area; and 6, classifying and identifying the extracted features, and judging whether hand gesture belongs to the illegal driving behavior. By the method, the illegal behavior of a driver in a driving period can be effectively determined, and a traffic accident which is caused by the illegal operation of the driver can be effectively avoided.

The invention provides a multi-feature-fusion-convolutional-neural-network-based plankton image classification method. The method comprises: lots of clear plankton images are collected and a large-scale multi-type plankton image data set is constructed; a global feature and a local feature are extracted by using an image conversion and edge extraction algorithm; an original feature image, a global feature image, and a local feature image are inputted into a depth-learning multi-feature-fusion convolutional neural network to carry out training, thereby obtaining a multi-feature-fusion convolutional neural network model; and then the plankton images are inputted into the multi-feature-fusion convolutional neural network model and classification is carried out based on a finally outputted probability score. According to the invention, the angle of biological morphology, the computer vision method, and the depth learning technology are combined; and thus the classification accuracy for plankton images, especially large-scale multi-type plankton images is high.

The invention discloses a high-speed weak target flight path detection method of an image field, relating to the field of moving object tracking, and mainly solving the problem of extracting a moving object flight path and a target kinematic parameter from data with a low signal-to-noise ratio. The detection process is as follows: firstly, obtaining a time-Doppler image by mixing and accumulating target multiframe data; carrying out edge extracting on the basis of phase equalization model; carrying out skeletonization and binary image de-noising treatment on the edge detection result with a morphology method to realize edge center positioning as well as noise inhibition; and finally, using a least square method to detect the target flight path and estimate a target parameter. The invention can accurately estimate the advantages of a curve flight path parameter, and can be used for the data processing field for monitoring systems, such as radars and the like to realize high-speed weak target detection and tracking.

The invention discloses a lane line detection method based on feature point piecewise linear fitting. The method mainly includes the following steps that image reading is conducted; region-of-interest region reading is conducted on a read image; edge extraction is conducted on an extracted region-of-interest region image; binaryzation is conducted on an edge image of an interest road; feature points are extracted from the binarized edge image, and a candidate feature point vector set is obtained; filtering is conducted on the candidate feature point vector set, and filtered feature point vectors are obtained; piecewise linear fitting is conducted on the filtered feature point vectors by using the least square method, and candidate lane line equations are obtained; the candidate lane line equations are recognized; according to features of video frames, recognized lane lines are tracked, and a final lane line is output. According to the method, lane lines of lanes can be effectively detected, and the method has good anti-jamming capability and good fault tolerance. Meanwhile, the method is small in computation amount, simple to achieve and high in practicability.

A background subtraction apparatus of the present invention includes a key point locator for locating key points on a known object type, a boundary point locator for locating boundary points of the known object that make up the edges of the known object, and an edge processor for processing the edges to provide a clean-edged extraction of the known object from a background image. Preferably, the key point locator includes an alignment detector for detecting alignment of an image of the known object type with a skeleton image. Still more preferably, the skeleton image is an exoskeleton image and the known object type is a human being.

A vehicle includes: an image acquisition means (4) which acquires a road image via an imaging means (3); an edge extraction means (5) which extracts edge points from the acquired image; a line search means (6) which searches the extracted edge points for line components; an evaluation value calculation means (7) which calculates an evaluation value indicating the degree that each line component approximates a linear lane mark on the road for the line components searched for; a lane mark candidate selection means (8) which selects line components each having an evaluation value greater than a predetermined threshold value as candidates for a line component corresponding to the lane mark from the line components searched for; and a lane mark detection means (9) which detects the lane mark by determining the line component corresponding to the lane mark from the selected candidates for the line component. This allows the detection accuracy to be increased by preventing an object other than a lane mark on the road from being incorrectly detected as a linear lane mark such as a white line when detecting a linear lane mark such as a white line from the road image.

The invention pertains to the technical field of machine vision and provides a method for automatically extracting elliptic image features from a complex background image. In the invention, firstly a Canny edge extraction method is utilized to obtain the sub-pixel image coordinates of an image edge, and the adjacent edge pixels are connected to an edge fragment via an edge connection method; arc fragments in the edge fragments which meet constraints of the central distance and the constraints of the radius are fused through using an edge split fusion algorithm based on the arc approximate fit; the arc fragments which pertain to the same ellipse are clustered on the basis of a random sampling consistency robust estimation algorithm; and finally an elliptic parameter equation is obtained by carrying out the least square fitting of the clustered arc fragments in the same category. The proposed elliptic automatic extraction algorithm effectively eliminates the interference of the noise edge in the image, solves the problems of identification and clustering of the elliptic edge of the complex background image, and realizes the automatic extraction of the elliptic image features.

This invention relates to the field of object volume survey, which discloses a method based on the image volume metering and the measuring technique. This metering equipment comprises the image gathering installment and the computer. The image collecting installment comprises objective table, as well as three cameras established symmetrically on objective table; Take the above metering equipment as the foundation, this measuring technique including the camera demarcation, image gathering, the image pretreatment, the picture edge extraction, the coordinate transformation, the three dimensional restructuring step, finally counts all three dimensional spots on the measured object, then obtain the actual volume of objects. This invention may be widely applied in areas such as medicine, chemical industry, food, environmental protection, water treatment, building materials, grain micro powdery, the granulated material continual quota throws in Canadian, ingredient craft domains dried noodles batch feeding and so on.

A high-precision small modulus gear vision measurement method is disclosed. The method is characterized by after a CCD camera is calibrated, carrying out small modulus gear coordinate conversion, outputting an actual physical coordinate of a small modulus gear and calculating a correlation parameter coordinate scale factor, a rotation matrix and a translation matrix; reflecting a space position relation between a gear to be measured and a measurement system in the measurement system; carrying out image processing, wherein a processed object is a small modulus gear edge contour image acquired from a CCD camera 3; and carrying out denoising; then using and carrying out 1 / 8 sub-pixel edge extraction, extracting an edge point coordinate, carrying out center fitting and acquiring a center coordinate of the image; according to an edge point, acquiring a basic gear parameter, wherein the basic parameter is a tooth number, a mode number, a reference circle diameter and a pressure angle; fitting a standard gear contour curve; and comparing an actually-acquired edge contour curve with a theoretical contour curve so as to acquire a pitch deviation and a tooth profile deviation. The basic gear parameter and quality can be comprehensively analyzed and evaluated.

The invention discloses an efficient DSP banknote serial number recognizing method, which relates to the technical field of image recognition technologies. The method of the invention mainly comprises the key steps of banknote edge extraction, facing direction recognition, serial number location, character segmentation and character recognition. Through banknote edge extraction, the banknote edge in an image is searched and fitted. Facing direction recognition is carried out to determine the front and the back of a banknote and judge whether the banknote is inverted so as to know the rough position of the serial number. Serial number location refers to locating the serial number coarsely based on prior knowledge, carrying out tilt correction on the basis and locating the serial number precisely using horizontal projection. Then, the characters of the precisely located serial number are segmented to obtain independent characters, and on the basis, the gray scale and size of the obtained segmented characters are normalized. Finally, the full-pixel characteristics of the character image are extracted, and the characters are identified using a neural network method.

An image processing unit is disclosed that smoothens a notice pixel of input image data with pixel values of pixels around the notice pixel to reduce noise in the image data. The image processing unit includes an edge extraction unit that extracts an edge based on the pixel values of the image data; a region determination unit that determines a region to be processed where a sum of edge intensities of the pixels including the notice pixel becomes greater than or equal to a threshold; and a smoothening unit that changes a smoothening intensity in accordance with a size of the region to be processed to smoothen the notice pixel corresponding to the region to be processed.

The invention discloses an image characteristic registration based geometrical fine correction method for an aviation multispectral remote sensing image, comprising following steps of: 1) utilizing a Sobel edge detection operator to carry out edge extraction on both an aviation multispectral remote sensing image to be registered and a normal incidence high-resolution satellite image which is taken as a standard; 2) utilizing a Harris angular point detection method to detect angular points of the aviation multispectral remote sensing image to be registered and the normal incidence high-resolution satellite image which is taken as the standard; 3) carrying out rough correlation between the two images through a correlation method; 4) carrying out fine correlation between the angular points of the two images through calculating supporting strength; 5) carrying out back calculation to obtain a multinomial coefficient according to a multinomial model; and 6) carrying out gray level re-sampling by adopting a bilinear interpolation to obtain a registered image. According to the image characteristic registration based geometrical fine correction method for the aviation multispectral remote sensing image, disclosed by the invention, the geometrical correction problem of the aviation multispectral remote sensing image lacking of a ground reference point can be better solved and the geometrical correction precision of an aviation multispectral scanner is improved, so that an aviation remote sensing technology can be better applied to production livings of national economy.

The invention discloses an image super-resolution reconstruction method and system based on edge detection, and the method comprises the steps: taking a high-resolution image as an original image sample, and obtaining a low-resolution image after the image preprocessing; constructing a super-resolution reconstruction network model, and inputting the low-resolution image into the super-resolution reconstruction network model to pre-train the super-resolution reconstruction network model; calculating the super-resolution image and the high-resolution image through a VGG19 network to obtain a perception loss function; carrying out color space conversion and edge extraction on the super-resolution image and the high-resolution image, and obtaining an edge loss function through L1 loss functioncalculation; and combining the perception loss function and the edge loss function in proportion to obtain a loss function Ltotal and update parameters, and inputting the low-resolution image into the trained super-resolution reconstruction network model to obtain a recovered high-resolution image. According to the invention, the defect of fake texture generation is overcome, and the visual effect and authenticity of image restoration are improved.

The invention discloses a binocular-vision-based rapid detection method for three-dimensional information of a space object. The method comprises: (1). Two cameras are calibrated to obtain an internal parameter matrix and an external parameter matrix; (2), the two cameras collect an image of a measured object simultaneously; (3), according to the calibration result, a re-projection matrix is calculated and three-dimensional correction is carried out on the image; (4), filter and smoothing processing is carried out on the corrected image and edge extraction is carried out on the filtered image; (5), on the basis of an improved edge-feature-based rapid matching algorithm, three-dimensional matching is carried out on the image to obtain a plurality of matching point pairs; and (6), according to the internal and external parameter matrixes obtained by camera calibration, the re-projection matrix, and a parallax principle, three-dimensional information of the measured object is restored. According to the invention, the traditional feature-point-based matching algorithm is improved. The original precision is kept and the matching speed is increased; and real-time three-dimensional information detection can be realized.

The invention discloses an image matching method based on polygon generalized Hough transform. The method includes: (1) off-line stage: (1.1) performing edge extraction and polygon fitting of a template image; (1.2) building an R-Table table by employing the local triangle characteristics and the vector relation between the vertex and a reference point of the template image; (2) on-line stage: (2.1) performing edge extraction and polygon fitting of a target image; and (2.2) voting for the target image; the position point with the highest vote is regarded as a matching point corresponding to the reference point of the template image so that the corresponding relation between the vertex of the polygon of the template image and the vertex of the polygon of the target image is determined, and parameters for displacement, rotation and zooming between the template image and the target image are determined. The method is used for real-time positioning of objects in RFID manufacturing equipment and LED manufacturing equipment, and the robustness and the precision are good.

To provide a motion vector detection device capable of rapidly and accurately detecting a main motion vector from plot data on a computer screen. An edge extraction unit (201) extracts edge points from each of a previous frame and a current frame. A feature point extraction unit (202) extracts, as a feature point, an edge point which is uniquely determined based on a relative positional relationship with another at least one edge point within a frame to which the edge points belong, from among a plurality of extracted edge points. A feature point pair extraction unit (203) extracts, as a feature point pair, a pair of feature points having a common positional relationship, among pairs of a feature point of the previous frame and a feature point of the current frame. A motion vector calculation unit (240) calculates, as a motion vector, a difference between coordinate values of the feature point pair extracted from the previous frame and the current frame.