68 results about "Feature orientation" patented technology

The present invention provides a text implication relation recognition method which fuses multi-granularity information, and proposes a modeling method which fuses multi-granularity information fusionand interaction between words and words, words and words, words and sentences. The invention firstly uses convolution neural network and Highway network layer in character vector layer to establish word vector model based on character level, and splices with word vector pre-trained by GloVe; Then the sentence modeling layer uses two-way long-short time memory network to model the word vector of fused word granularity, and then interacts and matches the text pairs through the sentence matching layer to fuse the attention mechanism, finally obtains the category through the integration classification layer; After the model is established, the model is trained and tested to obtain the text implication recognition and classification results of the test samples. This hierarchical structure method which combines the multi-granularity information of words, words and sentences combines the advantages of shallow feature location and deep feature learning in the model, so as to further improve the accuracy of text implication relationship recognition.

The invention provides a pedestrian inertial positioning system based on indoor magnetic field feature assistance. The system comprises a magnetic field and inertial data obtaining module, a magnetic field positioning module, a pedestrian dead reckoning module, a positioning fusion module and an output module, wherein the magnetic field and inertial data obtaining module is used for acquiring magnetic field, accelerated speed and angular velocity information; the magnetic field positioning module is used for building a magnetic field feature library and carrying out time-frequency analysis on the magnetic field vector sequence in real time to extract the time-frequency feature, and matching with the magnetic field feature library to carry out magnetic field feature positioning; the pedestrian dead reckoning module is used for updating accelerated speed and angular velocity zero offset according to the condition that the step velocity discontinuity is zero during walking, judging the step number and calculating the step length and the direction of each step; the positioning fusion module is used for fusing a magnetic field feature positioning result and a pedestrian dead reckoning inertial positioning result by means of particle filter; and the output module is used for displaying a positioning result on web pages and terminals. The system provided by the invention has the characteristics of being independent from beacon during positioning, low in cost and consumption of positioning terminals, accurate in positioning result and adaptive to environment change.

An image processing technique which identifies pixels in images which are associated with features having a selected shape, such as but not exclusively step edge, roof, ridge or valley. The shape of the intensity profile in the image is compared in an intensity independent way with a shape model to select those pixels which satisfy the shape model and are thus associated with the feature of interest. This comparison is achieved by examining the phase and amplitude of a spectral decomposition of parts of the image profile in the spatial or spatio temporal frequency domain. This decomposition can be achieved using quadrature wavelet pairs such as log-Gabor wavelets. The difference between the odd and even components, known as the feature asymmetry, gives an indication of the shape of the feature. The analysis may be extended to the time domain by looking at the shape of the image profile across a time sequence of images, which gives an indication of the velocity of a moving feature. Pixels identified as belonging to a feature of the right shape are labelled with the value of feature asymmetry, the local amplitude, feature orientation and feature velocity, and this information can be used to improve the tracking of detected features through a sequence of images.

An optical reduction system for use in the photolithographic manufacture of semiconductor devices having one or more quarter-wave plates operating near the long conjugate end. A quarter-wave plate after the reticle provides linearly polarized light at or near the beamsplitter. A quarter-wave plate before the reticle provides circularly polarized or generally unpolarized light at or near the reticle. Additional quarter-wave plates are used to further reduce transmission loss and asymmetries from feature orientation. The optical reduction system provides a relatively high numerical aperture of 0.7 capable of patterning features smaller than 0.25 microns over a 26 mm×5 mm field. The optical reduction system is thereby well adapted to a step and scan microlithographic exposure tool as used in semiconductor manufacturing. Several other embodiments combine elements of different refracting power to widen the spectral bandwidth which can be achieved.

The present invention discloses a video facial attaching picture effect processing method and generation system and belongs to the video effect field. The invention aims to make a user edit the effectof an attaching picture and adjust the position and size of the attaching picture by himself or herself. The method comprises the following steps that: feature positioning is performed on a face in avideo so as to obtain the facial feature points of the face; a selected image is imported and added on the face in the video so as to be adopted as an attaching picture, wherein the attaching pictureincludes an attaching picture image and attaching picture control points, the position parameters of the attaching picture control points are calculated according to the attaching picture control points and the facial feature points of the face which are corresponding to the attaching picture control points, and transformation processing is performed on the attaching picture through the positionparameters of the attaching picture control points, and the position parameters of the attaching picture control points can identify a relative position relationship between the attaching picture anda specified attaching picture area of the face in the video; and the attaching picture which has been subjected to the transformation processing is added onto the specified attaching picture area of the face in the video. The structure of the system includes a display module, a function module and a prompting module. With the video facial attaching picture effect processing method and generation system adopted, a user can edit the effect, position and size of an attaching picture by himself or herself.

Laser-enhanced visual simultaneous localization and mapping (SLAM) is disclosed. A laser line is generated, the laser line being incident on an object and / or environment. While the laser line is incident on the object, one or more images of the object with the laser line incident on the object are captured. The camera is localized based on one or more characteristics of the laser line incident on the object. In some examples, improved feature localization provided by the laser line provides more accurate camera localization, which, in turn, improves the accuracy of the stitched mesh of the object / environment. As such, the examples of the disclosure provide for improved camera localization and improved three-dimensional mapping.

The invention discloses a method applied to palm shape extraction and feature positioning in a high-freedom degree palm image, which can be suitable for the palm shape extraction and the feature positioning in the high-freedom degree palm image and has favorable adaptability to the classes, the placing positions and the angles of left and right hands, and palm shapes and sizes and finger opening and closing degrees in the palm image. Based on the distance of lower edge points of palm-shaped corrective gravity-center polar coordinates and the distribution regularity of angles, the method realizes the positioning of finger pits of finger tips of the palm-shaped zone in the palm image so as to judge the left and the right hand class of the extracted palm and calculate the rotating angle of the palm. The output results of the method are the integral palm shape and the positioning characteristic point of the palm image, users can conveniently complete the positioning and the extraction of various characteristics comprising the palm shape, palm prints, fingerprints and the like of a palm sample through the data, and therefore, the method has universality. Through the test of self-established high-freedom degree palm image library, the result indicates that the method has high precision rate and robustness.

The invention discloses a human face feature positioning method based on an ASM algorithm, belongs to the technical field of image processing, and mainly relates to human face recognition technology in biological feature authentication. The method comprises the following steps: firstly, performing manual calibration on a group of human face sample images to extract feature points; secondly, adopting a Procrustes Analysis algorithm to carry out registration on a sample set to obtain an average human face model; thirdly, adopting a strategy of performing contour search on the images under three resolution ratios to establish a statistic grayscale search model; and finally, performing a matching operation on the statistic grayscale search model and a local grayscale model of a human face image Ys to be positioned, and performing cyclic iterative search positioning on the human face image Ys to be positioned. The human face feature positioning method combines the ASM algorithm and the Procrustes analysis method, can effectively improve human face positioning speed and accuracy, and has strong commonality.

A high-accuracy optical aberration correction system and method. Z-heights determined by an auto-focus tool, which would otherwise vary depending on the orientation angle of surface features or edges in the focus region of interest, and on the location of the focus region of interest in the field of view, are corrected based on a novel error calibration method. Error calibration data includes a set of Z-corrections for a range of different feature orientation angles (e.g., 180 degrees), for multiple locations in the field of view. Error calibration data may be interpolated to correspond to the location of the current focus region of interest. During auto-focus measurements, a Z-correction may be adapted for a current measurement by weighting orientation-dependent error calibration data based on a histogram of the gradient (edge) directions present in the current focus region of interest.

A facial feature analysis system is provided. The disclosed system includes a virtual filter bank and a virtual discriminator. The virtual filter bank comprises a feature localization main module having an ancillary data bank which supplies a signal Sg for control functions. The system calculates face localization based on parameters of a holistic face-model, calculates feature localization based on parameters of an adaptive face graph, calculates feature extraction using stored feature values corresponding to selected validation and provides output signals using a signal delivery main module controlled by static and dynamic classification. The virtual discriminator bank calculates a user adapted allocation based on a face feature of the user and provides a periphery allocation for at least one command modus. The system may also be employed for providing execution signals for a manipulator, door surveillance, alarm systems and control of a vehicle.

The invention provides a method and a device for positioning human face features. The corresponding method comprises the following steps of: acquiring initial human face shapes through a preset human face detection algorithm, wherein the initial human face shapes comprise the width and the length of human faces and the central positions of the human faces; searching the initial human face shapes through decision functions to acquire human face feature points; determining whether the human face feature points are accurately positioned or not through algorithms of preset iterations and preset convergence probability; searching the initial human face shapes again through the decision functions if the human face feature points are not accurately positioned; and outputting the positioned human face feature points if the human face feature points are accurately positioned. According to the method and the device, the human face feature points are judged through the iteration algorithm and the convergence probability, so that whether the human face feature points are accurately acquired or not is determined, the human face features can jump out of the algorithm and are reacquired when not accurate, the searching time and the calculated amount can be reduced, and the identification accuracy of the human face feature points is further improved.

The invention discloses a robot vision inertial point linear feature positioning method and device. The robot vision inertial point linear feature positioning method comprises the steps that a priorithree-dimensional map of a current scene and measuring data of an inertial sensor are obtained, and the priori three-dimensional map is constructed in advance and comprises the three-dimensional pointlinear feature; and a current image of a robot and measuring data of an inertial sensor are obtained. According to the robot vision inertial point linear feature positioning method and device, fusionof multiple sensors is utilized, advantage information of all the sensors are fully exerted in an algorithm, and the effect of improving the position accuracy and robustness is finally achieved.

A method for manufacturing a tooth implantation plate includes such steps as using CT to obtain scanned oral cavity image data as reference object, copying a stereo model of oral cavity to obtain the shape of oral cavity, scanning the stereo model to obtain the model data of oral cavity as locating object, taking at least one relative characteristic from said reference object and locating object, locating the characteristic for locating said model data in said scanned image data, creating a digital model, using CAM to plan a machining route to obtain a digital manufacture model, and manufacturing tooth implantation plate.

A fine-grained behavior model matching based method and system for effective feature location in large legacy systems. A feature behavior analyzer extracts internal behaviors of a feature under requesting based on NLP techniques or writing rules of the feature specification and generates a feature behavior model. A method uses multiple information sources associated with each method under analyzing to generate an expressive behavior signature for each method. The method integrates control flow information of each method and the signature information of its callees, and generates a CFG-based behavior model. A feature location identifier identifies and ranks the feature-related code locations based on a similarity between the feature behavior and the code behavior models. In one aspect, “use cases”, “source code repository” and “issue tracking system” are historical information of existing applications that are used to help understand each code unit of legacy systems applications, and recommend code units related with the given feature description.

The invention discloses a feature localization method based on multi-source software data analysis in the technical field of software engineering, and aims to solve the technical problems that feature localization results are inaccurate and incomplete in the prior art. The method combines information retrieval technology, data mining technology and dynamic analysis technology to respectively perform feature localization on a current software system, a evoluting history library and executing tracks, and intersection calculation is performed on feature localization results by the three technologies to acquire final feature localization results, feature localization based on multi-source software data analysis is implemented, and better accurateness, complement and efficiency are provided; the three technologies are high in maturity, so that the method is easy to operate and implement. The method can be applied in feature localization in class hierarchy and method hierarchy, reality of the analysis can be combined, corresponding granularity hierarchy can be selected to perform feature localization, and a flexible selecting frame is provided for realistic multi granularity hierarchical feature location.

The invention discloses a method for manufacturing an artistic punched decorative plate by adopting characteristic-based hole layout design. The method comprises the following steps of (1) importing a color custom image of a hole decorative plate to a system and converting the color custom image into a grayscale image; (2) performing characteristic segmentation on the grayscale image; (3) performing characteristic inner hole layout: performing characteristic locating and characteristic classification, and selecting and using a layout method for different holes according to classification of characteristics; (4) performing boundary hole layout: performing characteristic reconstruction, cutting a characteristic boundary region, and performing characteristic boundary layout; (5) integrating process information to generate complete manufacturing process information of the artistic punched decorative plate; and (6) importing the manufacturing process information to a numerical control machining equipment system for performing a numerical control punching machining process on the artistic punched decorative plate. Variable-direction, variable-spacing, variable-density and variable-aperture multiple variable combination hole layout modes are realized; the imaging quality and the imaging effect are improved; and the imaging capability and the product quality of the hole decorative plate are optimized.

The invention relates to a method for manufacturing a dental implantation surgical guide plate. The method comprises the following steps of: planning a dental implantation; obtaining a digital plaster model of an oral cavity; providing dental implantation planning data for the digital plaster model by using feature positioning; then integrating both the digital plaster model and the dental implantation planning data into digital processing data; fixing a model material block to a processing position so that the digital processing data guides a processing machine to process the model material block into a solid oral cavity model which has a shape in fit with that of an oral cavity of a patient and has structures such as teeth, gingiva and dental implantation positions; arranging auxiliary positioning pieces at the dental implantation positions; and finally molding the surgical guide plate provided with a plurality of guide holes on the solid oral cavity model by matching a thermoplastic material with a rollover molding method or a vacuum molding method, and thus, the more accurate surgical guide plate can be manufactured under the condition of time saving, labor saving and low manufacturing cost.

The invention relates to a feature-oriented high-precision gravity map construction and interpolation method, which is characterized by comprising the following steps of: step 1, constructing a high-precision gravity map containing accurate feature information according to distribution characteristics of a gravity field feature; and step 2, calculating a gravity abnormal value at any point P(r,theta) in a region by using a multiple linear interpolation method or a radial gravity abnormal nonlinear feature equivalent method according to the high-precision gravity map constructed in the step 1. The feature-oriented high-precision gravity map construction and interpolation method adopts a direct description manner of the gravity field feature, which comprises a region gravity abnormalpeak point and nonlinear information falling along the radial gravity anomaly from the peak point, so that the description of the gravity field feature is more efficient and accurate, meanwhile, the shipborne measurement workload of constructing the gravity map is reduced, and the demand of the gravity field feature positioning technology on gravity field feature information is met.

An optical reduction system for use in the photolithographic manufacture of semiconductor devices having one or more quarter-wave plates operating near the long conjugate end. A quarter-wave plate after the reticle provides linearly polarized light at or near the beamsplitter. A quarter-wave plate before the reticle provides circularly polarized or generally unpolarized light at or near the reticle. Additional quarter-wave plates are used to further reduce transmission loss and asymmetries from feature orientation. The optical reduction system provides a relatively high numerical aperture of 0.7 capable of patterning features smaller than 0.25 microns over a 26 mm×5 mm field. The optical reduction system is thereby well adapted to a step and scan microlithographic exposure tool as used in semiconductor manufacturing. Several other embodiments combine elements of different refracting power to widen the spectral bandwidth which can be achieved.

The invention relates to a non-contact cheongsam customized human body three-dimensional size measurement method based on deep learning. The method comprises the following steps: firstly, carrying outtraining on a COCO data set to obtain a network model capable of separating a human body from a complex background; secondly, acquiring back and side images of an object; positioning the human body coordinates and shading according to the network model and extracting a binary image with the human body object as the center; then, according to the performance characteristics of each part of the human body, positioning coordinates of each characteristic point to obtain a first size group; and dividing the body type of the target human body into a plurality of categories according to the coordinates of the feature points and the first size group, and obtaining the three-dimensional girth size of the human body by using different calculation formulas, namely, a second size group. The inventionprovides a novel armpit point, neck point and waist high point positioning method for garment customization, provides a non-contact measurement method for the sizes of the front waist section, the rear waist section, the front chest width, the back width and the like for cheongsam customization, and meanwhile, carries out fine adjustment on feature point positioning and girth calculation according to human body features.

The invention discloses a three-dimensional blind watermarking method for constructing watermark synchronous information by using one-ring neighborhood area of a vertex and Beamlet line features. The method comprises the following steps: selecting M non-overlapped one-ring neighborhoods from small to large according to the one-ring neighborhood area of the vertex of a three-dimensional model; performing independent tangent projection on the M one-ring neighborhoods, extracting features of M Beamlet lines, and determining embedded positions of M watermark positions; modifying vertex coordinates corresponding to the Beamlet line features, so that two grid edges connected with the vertex coordinates have equal length for embedding watermark information bits; and performing watermark detection without the original three-dimensional model. The method is an airspace blind watermarking method, and obviously improves the robustness against affine transformation attack while ensuring the transparency of the watermark; and the algorithm also has strong robustness against grid simplification, smoothing and attack of noise and the like.

The invention discloses a driver inattentive driving detection method and a driver inattentive driving detection system based on video face analysis. The driver inattentive driving detection method comprises the steps of: acquiring a video sequence of the head of a driver during the driving process; performing face detection on the video sequence to obtain a face image; performing feature positioning on the face image; mapping the positioned face features onto a three-dimensional model to obtain face attitude data of the driver; extracting mouth feature point information from the positioned face features, so as to obtain mouth shape data of the driver; and judging whether the driver drives inattentively according to the face attitude data and the mouth shape data of the driver. The driverinattentive driving detection method and the driver inattentive driving detection system can improve the reliability and precision of judging non-fatigue driving of the driver in various complicated scenes such as at night, having uneven illumination, wearing sunglasses and the like.

The invention discloses a traffic sign recognition method and device and a vehicle, wherein, the method comprises the following steps: obtaining an image to be recognized by collecting an image in front of the vehicle; locating a traffic sign area in the image to be recognized according to color features and shape features of the image to be recognized; and obtaining SIFT features of the traffic sign area, and recognizing traffic signs in the traffic sign area according to the SIFT features of the traffic sign area. According to the method of the invention, the accuracy and the real-time property of the traffic sign recognition can be greatly improved.

A fine-grained behavior model matching based method and system for effective feature location in large legacy systems. A feature behavior analyzer extracts internal behaviors of a feature under requesting based on NLP techniques or writing rules of the feature specification and generates a feature behavior model. A method uses multiple information sources associated with each method under analyzing to generate an expressive behavior signature for each method. The method integrates control flow information of each method and the signature information of its callees, and generates a CFG-based behavior model. A feature location identifier identifies and ranks the feature-related code locations based on a similarity between the feature behavior and the code behavior models. In one aspect, “use cases”, “source code repository” and “issue tracking system” are historical information of existing applications that are used to help understand each code unit of legacy systems applications, and recommend code units related with the given feature description.

The invention relates to the technical field of automatic control and computer vision, in particular to a fatigue driving detection method based on a neural network, which comprises the following steps: acquiring a face image in real time through a camera; carrying out image preprocessing on the acquired face image; inputting the face image after image preprocessing into a convolutional neural network model, performing face positioning and face feature positioning, and outputting a face feature image after face feature positioning; wherein the face features comprise eyes and a mouth; calculating a PERCLOS parameter fP, a blink frequency BF parameter fBF and a yawn parameter fyawn according to the state of each face feature in the face feature image; calculating a fatigue index according to the PERCLOS parameter fP, the blinking frequency BF parameter fBF, the yawn parameter fyawn and the corresponding weight; and comparing the fatigue index with an early warning threshold, If the fatigue index meets the early warning threshold, generating early warning prompt information. The problem that an existing fatigue driving detection technology is low in detection accuracy is solved.