48 results about "Point distribution model" patented technology

The invention relates to a key point location method. The key point location method includes the steps: inputting an image, utilizing a pre-trained full convolution network to obtain the response diagram of each key point of an object-of-interest in an image; based on the response diagram of each key point, utilizing a pre-trained point distribution model to obtain initial location of each key point; and based on a weighted constraint mean value drifting method, performing iteration adjusting on location of each key point, and finally obtaining the final location of each key point. The key point location method organically integrates the expression capability of data drive with the prior inference capability of model drive; the full convolution network used by the key point location method can effectively respond to rigidity and non-rigidity conversion of the object in the image; the point distribution model can effectively respond to the shielding situation in the image; and the weighted constraint mean value drifting can reasonably balance the effect of the full convolution network and the point distribution model, so that the robustness of key point location can be greatly improved.

The invention relates to a method for directly generating digitalized shoe trees by using foot shapes, solving the difficult problem of customizing shoe trees by measuring feet. The method comprises the following steps: (1) selecting shoe tree samples to form a shoe tree sample set; (2) designing gauge points, determining gauge points on each shoe tree sample and then acquiring the positions of the gauge points by using a scanner to obtain gauge point distribution models of the shoe tree samples; (3) aligning the gauge point distribution models of the samples; (4) generating statistic deformation models of the shoe trees; (5) calibrating the parameter of each camera and then acquiring a plurality of images of target foot shapes; (6) calculating the center of gravity of the foot shapes; (7) estimating the sizes of the foot shapes; (8) estimating the direction of the foot shapes; (9) changing individual vectors in the statistic deformation models of the shoe trees according to the plurality of images of the foot shapes until the projection of the models in each image is most identical with the true images of the foot shapes; and (10) outputting digitalized three-dimensional shoe tree models. The invention has the advantages of full automation, high speed and efficiency, simple use, low cost and the like.

Certain embodiments provide a computer apparatus operable to carry out a data processing method to position a set of anatomical landmarks in a three-dimensional image data set of a part or all of a patient, comprising: providing a trained supervised machine learning algorithm which has been trained to place each of the set of anatomical landmarks; applying the supervised machine learning algorithm to place the set of anatomical landmarks relative to the data set; providing a trained point distribution model, including a mean shape and a covariance matrix, wherein the mean shape represents locations of the set of landmarks in a variety of patients; and applying the point distribution model to the set of landmarks with the locations output from the supervised machine learning algorithm by: removing any landmarks whose locations have an uncertainty above a threshold with reference to the mean shape and covariance matrix; followed by: an optimisation of the locations of the remaining landmarks by maximising joint likelihood that a new shape, derived from linear combinations of eigenvectors of the covariance matrix, is plausible.

The invention discloses a facial expression feature extraction method based on a point distribution model. The method includes the specific steps of setting up the human body facial shape model through coordinates for marking human body facial image facial feature points, selecting relatively-ideal shape vectors as initial samples, enabling other vectors to correspond to the initial samples in shape till the difference between two adjacent average shape vectors is smaller than a certain value, and conducting calculation through a Gabor wavelet kernel function and conducting sampling to achieve fine mapping of the facial feature points of a set of feature points with different frequencies and phases. By means of the method, according to the facial expression feature extraction method based on the point distribution model, the elastic graph matching of human face recognition is conducted through a Gabor wavelet conversion coefficient, ideal facial expression features which are not affected by the samples are obtained, and the method has the obvious advantages for extracting features which are kept unchanged under uncorrelated deformation, insensitive to noise and high in category distinction degree.

The invention relates to a moving method and device for shot points in an obstacle area of a three-dimensional seismic acquisition and observation system, and the method comprises the steps: buildinga theoretical shot point distribution model representing the distribution of shot points according to the coordinate data of a theoretical observation system; constructing a grid control parameter according to a theoretical shot line distance and a theoretical shot point distance in the theoretical shot point distribution model; forming a polygon model representing the boundary of the obstacle area according to the boundary of the obstacle area, wherein the number of theoretical shot points covered in the polygon model is n; m vacancy grids are formed around the polygon model according to thegrid control parameters, and m is larger than or equal to n; and moving n shot points in the polygon model to n vacancy grids. According to the invention, the smoothness of the shot line and the uniformity of shot point placement are maintained, and the overall placement rationality of a plurality of shot points is improved.

The invention provides a heterogeneous network time domain interference coordination method and system based on user fairness. A heterogeneous network comprises a macro base station and at least one picocell base station within a coverage range of the macro base station. The method includes the steps that a Poisson point distribution model is adopted for modeling network deployment of the heterogeneous network to obtain a heterogeneous network model; the heterogeneous network model is adopted for respectively calculating network fairness indexes of the heterogeneous network and the system capacity in a unit geographic area; a multi-target planning model is built, the reception reference signal measurement offset and the time domain almost-blank subframe density of the heterogeneous network model serve as variables, and the network fairness indexes and the system capacity in the unit geographic area serve as objective functions; an optimal solution of the multi-target planning model is calculated, the reception reference signal measurement offset and the time domain almost-blank subframe density is set according to the optimal solution. By means of the heterogeneous network time domain interference coordination method and system, the fairness of wireless network resources of all users can be guaranteed, and meanwhile the total system capacity of the stacked heterogeneous network can be improved.

The invention discloses an aortic contour segmentation algorithm based on an active shape model, which mainly overcomes the blurring or discontinuity of the aortic edge caused by the traditional method through the shape constraint, avoids the interference of other tissues, and realizes the accurate segmentation of the aortic arch part and the descending aorta part contour. The realization processis as follows: (1) marking feature points of a training set sample and performing normalizing and registration treatment on the shape vectors to establish a point distribution model; (2) sampling thegray level along the normal direction of the contour boundary of the training sample set to construct the gray level texture model of the training set; (3) searching for the best matching point of each marker point iteratively, and then uniformly performing constraining by the active shape model until the aortic contour converges.

The invention relates to a high frequency data real-time collection and transmission method for road network level bridges. Compared with the prior art, the method has the advantage that the defect that high frequency data cannot be preferentially transmitted in real time with pertinence is solved. The method comprises the following steps of establishing a bridge monitoring point distribution model based on monitoring elements; performing value assignment on monitoring priorities of the monitoring elements; dynamically calculating data reporting channels; and transmitting data in real time anddynamically reporting and transmitting the data in real time according to the established data transmission channels. According to the method, the priority division is carried out on the reporting ofsensor data collected at high frequencies of the bridges, the important key collection data information is transmitted preferentially, the data reporting channels are dynamically calculated, the idletransmission channels with high bandwidths are obtained, and the key collection data information is uploaded through the idle channels, thereby realizing real-time transmission of the high frequencydata.

An image aberration analysis method based on invariant moments is disclosed, comprising the steps of: 1) selecting proper moment invariable, and in a dynamic structure model, using point distribution model for expressing the object outline in the image; 2) active detection method of the model aberration: 2.1) for PDM in the dynamic structure model, wherein each marking point includes the corresponding weight, decomposing one PDM to a sub-image according to the weight; 2.2) computing the border moment invariable of each sub-image after training and gathering each sub-image and the models to deform in the set range; then computing the convergence central values of these border moment invariables; 2.3) the aberration detection for the model comprising the steps of: (1) partitioning the deformed models according to the predefined partitioning strategy; (2) computing the border moment invariable of each sub-image; (3) computing the aberration possibility of each sub-image in light of the formula (2), (3) and (4); (3) active correction method for the model aberration. The invention can efficiently detect and correct the aberration in the process of search, including the outstanding reliability.

The invention relates to a method for ship collision avoidance in open waters, and the method comprises the following steps of establishing a collision circle according to the speed and distance of two ships, wherein the collision circle refers to a set of all possible collision points of the two ships; extending the heading lines of the two ships, and taking the points where the two heading linesintersect the collision circle as the possible collision points; making a circle respectively by taking all the possible collision points as centers and setting the distance parameter as the radius to form a possible collision point distribution model of the circle; and making the heading line of one ship away from the possible collision point distribution model of another ship. The method provided by the invention is capable of avoiding complex function calculations.

The invention discloses an ASM-based lane line detection method, which comprises the steps of separating regions of interest from an acquired image ahead of a vehicle; pre-processing the image, including graying the image after the separation of the regions of interest, then carrying out median filtering by using a 3x3 sliding window, retaining the details and removing interference noises; calibrating feature points on two lane lines in the pre-processed image and establishing a training set formed by a feature point distribution model; carrying out normalization processing for the training set established in the third step and obtaining an aligned shape; capturing statistical information of the aligned shape through principal component analysis calculation, and establishing an ASM model; and acquiring to-be-detected images ahead of the vehicle in real time, separating regions of interest from the acquired to-be-detected images ahead of the vehicle and pre-processing the images according to the first and second steps, finally searching for information matching the ASM model from the pre-processed to-be-detected images ahead of the vehicle acquired in real time by using the established ASM model, and finishing the lane line detection.

The invention discloses a comprehensive construction method for an engine shell vibration general picture, and is used for solving the technical problem of poor practicality of an existing construction method for the engine shell vibration general picture. The technical scheme is characterized in that a way of uniform revolving speed and steady state clamping is adopted, and the vibration information of tested equipment in each area is measured in different times; then, the position information of each measurement point on the shell is synthesized to carry out compensation calculation on the original vibration information obtained by measurement, and the 1 / 3-time frequency interval vibration numerical value of each measurement point under a certain center frequency is calculated by the data; according to the outline structure of the shell, a three-dimensional model is established, and a measurement point distribution model is established according to a practical distribution situation; and finally, a measurement point coordinate and a measurement sequence are followed to splice a plurality of area surface vibration matrixes, so that the vibration information which is in the presence is converted into a matrix which can integrally express shell vibration characteristics, and data in the matrix corresponds to the measurement point of a shell model one by one to establish a three-dimensional vibration general picture model. The method exhibits good practicality.

The invention discloses a fair resource distribution strategy based on a light-weight content central network framework. The strategy comprises the steps of providing a binary tree node distribution model in scenes (such as airports and schoolyards) with relatively simple communication geographical environments; aiming at the binary tree node model, providing a fair resource distribution model fordistributing cache content; in order to solve the resource distribution model, providing a 0-1 integrated planning algorithm with a weight; and providing three indexes such as a hit rate, a network delay and a hit rate variance to assess reasonability and fairness of resource distribution.

The invention discloses a transmission network planning method and system. The method comprises the following steps: collecting a physical material of a transmission node, drawing a transmission nodedistribution model including two convergence points, and performing gridding processing and planning area division on the transmission node distribution model; determining a node parameter according to the transmission node distribution model after the planning area division, and determining a network topological structure according to the node parameter. According to the method in the invention,through the networking planning of performing gridding processing on the existing transmission node, the effect of best performance / lowest engineering cost can be reached according to actual needs, the intelligence of the transmission network planning is realized, and the labor cost is reduced.

The invention provides a moving target selection method and a moving target selection system based on correction of a user presentation model. The method comprises the steps of performing dimension matching on a display interface, wherein an optional target included in the interface is expressed by a target set; acquiring a selection fall point, and calculating a selected probability of each optional target in the target set according to the selection fall point to obtain a probability set; taking the optional target corresponding to a maximum probability in the probability set as a final selected target. According to the moving target selection method and the moving target selection system based on correction of the user presentation model, for a moving target selection task, a target selected fall point distribution model is established, and the moving target selection is further corrected through the model to improve the efficiency of interactive tasks.

The present invention relates to a method for designing dental prosthesis, an apparatus for the same, and a recording medium for recording the same. According to the method, sore spots on which a patient may feel pain due to wearing a dental prosthesis are predicted based on a sore spot distribution model which has been modeled with a plurality of data, and the dental prosthesis is digitally designed based on the predicted results. Accordingly, it is possible to improve the prediction accuracy of sore spots and rapidly design the prosthesis which is more appropriate for the patient.

A site distribution model identifies a set of sites to satisfy operational requirements of a clinical trial. An objective function is defined as: a first element indicating whether a site is included in the clinical trial; and a second element indicating whether a country is included in the clinical trial. There is a set of constraints including that an estimated total enrollment reaches a defined target enrollment. Computer code is generated to implement a site distribution model, using an optimization modeling language, based on the objective function and the primary set of constraints. The site distribution model is solved, if possible, to produce values of the site decision variable and the country decision variable. Otherwise, it is indicated to a user that a solution is not possible. If solving the site distribution model is possible, a list of clinical trial sites is produced from the site decision variable.

The invention aims to provide a backlight light guide plate brightness optimization method, and relates to the technical field of light guide plates. The method comprises the following steps: constructing a first optical model according to structural parameters of the light guide plate, and performing simulation through the first optical model to determine an arc cutting structure; and constructing a second optical model according to the light guide plate structure parameters and the arc cutting structure, and performing simulation through the second optical model to determine a dot distribution model. According to the method, the first optical model and the second optical model are constructed in stages, the arc cutting structure and the lattice point distribution model are optimized respectively, systematic optimization of the brightness of the backlight light guide plate is achieved, and the problems that the light energy utilization rate of the light guide plate is low and the brightness is uneven are solved.

The invention relates to a method for calculating characteristic values and characteristic vectors of a flexible body point distribution model based on an interpolation algorithm, comprising the following steps: (1) loading flexible distribution models Ma and Mb of two adjacent times Ta and Tb, wherein Ma=<gamma a, phi a and lambada a>, Mb=<* b, phi b and lambada b>; (2) respectively calculating average shapes * a and * b, characteristic matrixes phi a and phi b and characteristic values lambada a and lambada b by the flexible body distribution models Ma and Mb to set the continuous change ofa flexible body from the Ta to the Tb; (2.1) calculating the characteristic value of a middle model; and (2.2) estimating a characteristic vector of the middle model. The method can form middle transition models which are distributed in real time and has high model accuracy and good practicality.

The invention relates to a method for directly generating digitalized shoe trees by using foot shapes, solving the difficult problem of customizing shoe trees by measuring feet. The method comprises the following steps: (1) selecting shoe tree samples to form a shoe tree sample set; (2) designing gauge points, determining gauge points on each shoe tree sample and then acquiring the positions of the gauge points by using a scanner to obtain gauge point distribution models of the shoe tree samples; (3) aligning the gauge point distribution models of the samples; (4) generating statistic deformation models of the shoe trees; (5) calibrating the parameter of each camera and then acquiring a plurality of images of target foot shapes; (6) calculating the center of gravity of the foot shapes; (7) estimating the sizes of the foot shapes; (8) estimating the direction of the foot shapes; (9) changing individual vectors in the statistic deformation models of the shoe trees according to the plurality of images of the foot shapes until the projection of the models in each image is most identical with the true images of the foot shapes; and (10) outputting digitalized three-dimensional shoe tree models. The invention has the advantages of full automation, high speed and efficiency, simple use, low cost and the like.

The invention discloses a heart image four-dimensional context segmentation method, equipment, a storage medium and a device. The method comprises the following steps: constructing a point distribution model in a three-dimensional activity shape model according to a four-dimensional heart image; constructing an initial heart shape according to the four-dimensional heart image; processing the initial shape of the heart according to the point distribution model; obtaining an initial shape under the time phase of the diastolic period; and segmenting the image under the diastolic phase according to the three-dimensional activity shape model and the initial shape under the diastolic phase to obtain a segmentation result under the current time phase, the diastolic phase being the time phase corresponding to the maximum volume, and so that based on diastolic phase segmentation, the segmentation accuracy of the heart image is improved; and taking the segmentation result under the current timephase as the initial shape of the to-be-segmented heart image when the current time phase is adjacent to the current time phase, performing analogy in sequence to obtain segmented images of all time phases, and segmenting the images of each time phase of the heart one by one based on the segmentation result under the current time phase, thereby improving the segmentation efficiency of the heart image.

The embodiment of the invention provides a parameter inference method and device, and relates to the technical field of prediction or optimization, and the parameter inference method comprises the steps: receiving a plurality of key points of a human body part, wherein the plurality of key points are used for reflecting the features of the human body part; and processing the plurality of key points by using a trained neural network point distribution model to obtain shape parameters and posture parameters corresponding to the human body part. According to the embodiment of the invention, the trained neural network point distribution model is utilized to obtain the shape parameter and the attitude parameter corresponding to the human body part according to the plurality of key points of thehuman body part, and compared with the prior art, the problems of large calculation amount and long PDM model establishment time can be improved.

The invention discloses a face tracking method and a camera. The method of the present invention includes: obtaining the initial position of the facial feature point of the first frame image in the image frame sequence collected; using the initial position of the facial feature point as the input of the convolutional neural network, and the output of the convolutional neural network The position probability heat map of the facial feature point, the position probability heat map of the facial feature point adopts a point distribution model to perform iterative regression processing, and obtains the pixel position of the facial feature point in the first frame image; using the The facial feature point is at the pixel position of the first frame image, and the initial position of the facial feature point of the second frame image in the image frame sequence is acquired to realize face tracking. The invention can realize alignment and tracking of human faces between frames, avoid instability and shaking of facial feature points between frames, and improve tracking accuracy and tracking speed.

The invention provides a user click success rate prediction method for moving targets in any shape, belongs to the field of computer graphical user interfaces. The method comprises the following steps: representing the moving targets by using a sub-target set by determining the moving targets of which the click success rate needs to be predicted; performing weighted combination on the Gaussian distribution of all the sub-target sets to obtain a drop point distribution model of the moving target; acquiring a probability density function of normal distribution corresponding to the moving targetaccording to the drop point distribution model of the moving target and the shape of the moving target; and acquiring a function expression of the click success rate of the moving targets according tothe probability density function of normal distribution, and predicting the click success rate of each moving target through the function expression.

Embodiments of the present application provide a parameter inference method and device, which relate to the technical field of prediction or optimization. The parameter inference method includes: receiving multiple key points of human body parts, wherein the multiple key points are used to reflect the human body parts features; using the trained neural network point distribution model to process the multiple key points to obtain shape parameters and posture parameters corresponding to the human body parts. In the embodiment of the present application, the trained neural network point distribution model can be used to obtain the corresponding shape parameters and posture parameters of the human body parts according to multiple key points of the human body parts. longer question.

The embodiment of the application provides a fracturing implementation deployment optimization method and an electronic device, and relates to the technical field of fracturing reconstruction. The method comprises the following steps: performing spatial division according to a borehole trajectory of a region to be analyzed and preset reconstruction information, and obtaining a plurality of sub-regions to be analyzed; obtaining a first coordinate of a model grid unit based on a three-dimensional reservoir sweet spot distribution model that has been constructed, and mapping the first coordinate to a current borehole local coordinate system by using a preset coordinate conversion algorithm to obtain a second coordinate of the model grid unit; obtaining a sweet spot index set corresponding to a current sub-region to be analyzed by position screening; and outputting fracturing implementation deployment information according to a plurality of sweet spot index sets corresponding to the region to be analyzed. By combining the borehole trajectory and the preset reconstruction information, the fracturing influence range is divided, and the three-dimensional reservoir sweet spot distribution model is mapped to the borehole local coordinate system, so that the analysis and effective identification of the recoverable potential of the far-well region fracturing are realized, and the accuracy of fracturing deployment is improved.

The embodiment of the invention provides a fracturing implementation deployment optimization method and electronic equipment, and relates to the technical field of fracturing transformation. The method comprises the steps of performing space division according to a well track and preset transformation information of a to-be-analyzed area, and obtaining a plurality of to-be-analyzed sub-areas; based on the constructed three-dimensional reservoir dessert distribution model, first coordinates of model grid units are obtained, the first coordinates are mapped to a current borehole local coordinate system through a preset coordinate conversion algorithm, and second coordinates of the model grid units are obtained; obtaining a dessert index set corresponding to the current to-be-analyzed sub-region through position screening; and according to the plurality of sweet spot index sets corresponding to the to-be-analyzed area, fracturing implementation deployment information is output. The fracturing influence range is divided by combining the well track and the preset transformation information, and the three-dimensional reservoir dessert distribution model is mapped to the local coordinate system of the well, so that analysis and effective identification of the fracturing exploitable potential of the far-well area are realized, and the accuracy of fracturing deployment is improved.

The present application relates to a kind of material surface icing adhesion strength envelope line measurement experimental design methods, it includes the following steps: preparing target ice layer on the substrate to be measured;Obtain uniaxial tensile adhesion strength and uniaxial shear adhesion strength;In shear-tensile coupling state, the horizontal coordinate value and the vertical coordinate value of N first measured data points in the target ice layer and the interface adhesion strength envelope line of the substrate to be measured are obtained;Data fitting is carried out by inflection point-free distribution model and first inflection point distribution model, and the type of the interface adhesion strength envelope line of the target ice layer and substrate material is judged according to the goodness of fit of two distribution models;If it is inflection point-free envelope line, the inflection point-free curve corresponding to the uniform distribution model after fitting is used as the measured adhesion strength envelope line;If it is inflection point envelope line, again preset N second measured data points;And all data points are fitted using the second inflection point distribution model, and the measured envelope line is obtained.

The present application relates to a kind of static ice and the interface adhesion strength envelope measurement method of measured substrate, it includes the following steps: by measuring system obtains uniaxial tensile adhesion strength and uniaxial shear adhesion strength;N first measured data points are preset with horizontal coordinate obeying uniform distribution, and its horizontal and vertical coordinate value is calculated, then data fitting is carried out using inflection point-free distribution model and first inflection point distribution model, and the type of static ice and the interface adhesion strength envelope of measured substrate is judged according to two fitting degrees, if there is inflection point, the horizontal and vertical coordinates of the preset N second measured data points are calculated in combination with specified second inflection point distribution model;Then data fitting is carried out using the second inflection point distribution model to all data points, and the adhesion strength envelope of measured substrate is obtained.