51 results about "Outlier removal" patented technology

A system for measuring a pulse and respiratory rate from passive thermal video includes contour segmentation and tracking, clustering of informative pixels of interests, and robust dominant frequency component estimation. Contour segmentation is used to locate a blood vessel region to measure, after which all pixels in the nearby region are aligned across frames based on the segmentation's position, and scale in each frame. Spatial filtering is then performed to remove noise not related to heart beat and then non-linear filtering is performed on the temporal signal corresponding to each aligned pixel. The signal spectrum of each pixel is then feed to a clustering algorithm for outlier removal. Pixels in the largest cluster are then used to vote for the dominant frequency, and the median of the dominant frequency is output as the pulse rate.

The present invention discloses a three-dimensional reconstruction method and system capable of maintaining sharp features. The method includes the following steps that: 1) a two-dimensional image filtering method extended to a three-dimensional space is adopted to perform smoothing de-noising on inputted roughly-registered point cloud; 2) an improved region growth method is adopted to perform outlier removal on the smoothed roughly-registered point cloud; 3) a kd-tree (k-dimensional tree) acceleration-based ICP (iterative closest point) algorithm is adopted to perform precise registration on the outlier-removed roughly-registered point cloud; 4) a neighborhood search and boundary point detection-based fusion method is adopted to fuse the precisely-registered point cloud; and 5) a feature point detection and adaptive step size update-based method is adopted to perform surface reconstruction on the fused precisely-registered point cloud. The three-dimensional reconstruction system is composed of a point cloud preprocessing module, a point cloud combining module and a surface reconstruction module. The three-dimensional reconstruction system which is realized based on the method of the invention can maintain the sharp features of the edge of a reconstructed model, and therefore, reconstruction speed is considered with accuracy ensured.

A correlation relationship between two metric time series is determined after removing the impact of outlying metric values (“outliers”) that are unimportant for analytical purposes. Each of the metric time series can represent values of different system metrics obtained by mining data gathered through the monitoring of cloud deployments. The outliers can be determined based on a maximum concentration interval of the data. Removing the impact of the outliers enhances the correlation of the metric time series and provides a better representation of the correlation relationship.

Enhanced removing of noise and outliers from one or more point sets generated by image-based 3D reconstruction techniques is provided. In accordance with the disclosure, input images and corresponding depth maps can be used to remove pixels that are geometrically and / or photometrically inconsistent with the colored surface implied by the input images. This allows standard surface reconstruction methods (such as Poisson surface reconstruction) to perform less smoothing and thus achieve higher quality surfaces with more features. In some implementations, the enhanced point-cloud noise removal in accordance with the disclosure can include computing per-view depth maps, and detecting and removing noisy points and outliers from each per-view point cloud by checking if points are consistent with the surface implied by the other input views.

A method is described for the automatic determination of a planogram in a vending machine using image processing. Novel steps include the use of a Gaussian map in a two-dimensional color space, such as the HS plane, to create high-dimensionally color vectors for all and for selected portions of images. Multiple feature detection / extraction algorithms are run between multiple idealized reference images for a product and one image from one vending machine coil location. The large resulting candidate feature list is pruned in a series of steps using both color and gray-scale color vectors and small area image matching around features. Remaining candidate features are ranked by a RANSAC outlier removal step, with the top ranked product then being the correct product in that coil in the planogram. Steps are repeated for all coils in a vending machine.

The invention provides a method for modeling a sea wave significant wave height inversion model based on particle swarm optimization (PSO) self-adaptive piecewise linear fitting, and belongs to the technical field of sea wave parameter inversion. The method comprises the steps of performing outlier removal processing on data, performing sparsification processing on the data, initializing parameters in a particle swarm, initializing a particle speed, updating the particle speed, updating particle displacement and the like. According to the method for modeling the sea wave significant wave height inversion model based on self-adaptive piecewise linear fitting, the wave height is subjected to inversion by utilizing a particle swarm optimization, so that the function of a conventional algorithm can be realized, the precision of the conventional algorithm is achieved, and more precise wave height inversion can be performed; and moreover, when the number of the pieces in the method is more than or equal to two, the modeling precision is higher than that of a conventional modeling method. The inversion model modeled by the method has higher inversion precision compared with the inversion model modeled by the conventional method, and moreover, the method for modeling a sea wave significant wave height inversion model based on PSO self-adaptive piecewise linear fitting is wide in applicability and high in flexibility.

The invention discloses a three-dimensional human head point cloud feature extraction method and a device thereof. The method comprises steps that 1, point cloud data pre-processing, sampling, outlier removal and noise removal are included; 2, training stage, a normal and an FPFH (fast point feature histogram) value of each point are calculated, and a K-dimensional tree is established; 3, FPFH values of three typical feature portions are segmented and extracted from a given model; 4, searching is carried on the K-dimensional tree established in the step 2, and a candidate set is searched; and 5, a principal curvature of each point of the candidate set and a shape response factor are calculated, a feature point is finally determined according to profiles of the three typical features and the position of a nose tip. Through the method, sensitivity to illumination and angles during two-dimensional feature extraction in the prior art is overcome, training of the large-scale data is not required, and rapid and accurate extraction of the three typical human head features is realized, and typical features of the nose tip, a left auditory meatus and a right auditory meatus can be extracted on the basis of the three-dimensional human head point cloud data.

Disclosed are a homography estimation apparatus and method. A down-sampling unit reduces the reference image and the target image to the same size to generate down-sampled images. A feature point extraction unit extracts feature points from the reference image and the target image down-sampled to the same size, respectively. An outlier removal unit matches the feature points extracted from the reference image with the feature points extracted from the target image, and detects and removes outliers, which are feature points that do not match, from the matched feature points. A homography estimation unit estimates a homography using feature points from which outliers are removed. A reference image correction unit corrects the reference image using the estimated homography.

Provided is an image processing apparatus. When a plurality of images acquired by photographing different directions is input, a region search unit of the image processing apparatus may search for an overlapping region within the plurality of images. An outlier removal unit may remove an outlier within the retrieved overlapping region.

A cell image segmentation method includes receiving a cell image, performing a nuclei initialization step to find an internal marker and an external marker to obtain a potential nuclei and a potential cell boundary, calculating a gradient map of the received cell image, performing a filtering step on the gradient map to generate a filtered gradient map, performing a nuclei detection step to obtain a segmented nuclei, and performing a nuclei validation step to obtain a valid nuclei. The nuclei initialization step includes performing a blob detection step to obtain a nuclei candidate, an outlier removal step to obtain the internal marker, a distance transform step to obtain a distance map, and a cell boundary initialization step to obtain the external marker. In another embodiment, a nuclear-to-cytoplasmic ratio evaluation method using the above cell image segmentation method is proposed.

The invention discloses a hot-rolled strip steel plate convexity prediction method based on deep learning, and the method comprises the following steps: S1, collecting and recording strip steel production data, and carrying out the preprocessing, including missing value deletion, abnormal value deletion and normalization, of the collected data; S2, according to the strip steel production data, removing redundant and irrelevant attributes in the collected data through an attribute selection method based on a Morita index, and screening a minimum number of attributes capable of representing plate convexity changes to form an input variable set of a forecasting model; and S3, establishing a strip steel outlet plate convexity prediction model based on the deep and wide convolutional neural network based on the input variable set so as to obtain the hot rolled strip steel outlet plate convexity. According to the method, high-order features and invariant features of the data are extracted byusing the convolutional layer in the convolutional neural network, the local correlation between the variables is learned, and the global feature learning ability of the deep neural network is combined, so that the plate convexity prediction precision is remarkably improved.

The invention belongs to the field of image recognition, particularly relates to an adversarial sample generation method, system and device for an outlier removal method, and aims to solve the problems that an adversarial sample adopted by existing classification model training based on deep learning cannot make an image classification error under an outlier removal method; and therefore, the trained classification model is poor in robustness and low in accuracy. The method comprises the following steps: acquiring a training data set with category labels, inputting three-dimensional point cloud data into a classification model, calculating classification loss, respectively calculating the gradient of the classification loss relative to the three-dimensional point cloud data and the gradient of the classification loss relative to outlier-removed three-dimensional point cloud data, and fusing the two gradients by multiplying a scaling factor to generate fusion disturbance, and applying the fusion disturbance to the three-dimensional point cloud data for repeated iteration to generate an adversarial sample. The generated adversarial samples can still cause image classification errorsunder the condition that outliers are removed, and the robustness and classification accuracy of the trained model are improved.

The invention discloses a delay estimation method and device used for indoor underwater target positioning. The device comprises a signal generating module, a signal acquiring and processing module, an ultrasonic transducer and hydrophones. The delay estimation method includes the steps that firstly, outlier removal and normalization are performed on acquired signal data; secondly, circulating relevant processing is performed on signals of the hydrophones and original signals of the transducer; thirdly, peak values of relevant signals are solved through a lead-lag method and a threshold method; if the two solved peak values are different, read-in data of a data acquisition card are wrong, data are re-read and used for calculation, and when the two solved peak values are the same, the peak values are corrected through a tracking envelop solving method, and final peak values and transverse coordinates corresponding to the peak values are obtained, and sound signal delay is obtained by multiplying transverse coordinate values by an AD sampling period. By means of the method and device, the delay estimation operation time can be shortened, accuracy is higher than the accuracy of a common cross-correlation method, and therefore underwater target positioning accuracy can be effectively improved.

The invention provides a data screening method and a data screening device. The data screening method comprises the steps of drawing a scatter diagram based on obtained discrete data in a to-be-screened data set; based on the distribution characteristics of each discrete point in the scatter diagram, determining a fitting curve connecting the plurality of discrete points and the fitting degree ofthe fitting curve; if the fitting degree of the fitting curve is smaller than a preset threshold value, constructing a probability distribution model based on the distance from each discrete point tothe fitting curve; determining a confidence interval range of the distance based on the average value and the standard error of the probability distribution model and the obtained saliency level value; and determining discrete data corresponding to the discrete points, of which the distance values with the fitting curve are located outside the confidence interval range, in all the discrete points,screening the determined discrete data from a to-be-screened data set, and determining the data set with the discrete data screened as a target data set. Thus, outliers can be removed with the fitting curve as the reference, and the outliers removal accuracy and thoroughness are improved.

Enhanced removing of noise and outliers from one or more point sets generated by image-based 3D reconstruction techniques is provided. In accordance with the disclosure, input images and corresponding depth maps can be used to remove pixels that are geometrically and / or photometrically inconsistent with the colored surface implied by the input images. This allows standard surface reconstruction methods (such as Poisson surface reconstruction) to perform less smoothing and thus achieve higher quality surfaces with more features. In some implementations, the enhanced point-cloud noise removal in accordance with the disclosure can include computing per-view depth maps, and detecting and removing noisy points and outliers from each per-view point cloud by checking if points are consistent with the surface implied by the other input views.

A computer implemented signal processing method to perform a cross-calculation between a first and a second signal by performing a cross-correlation for segments of said first signal with said second signal to obtain a plurality of partial cross correlation functions, obtaining a combined cross-correlation function by combining said partial cross-correlation functions to obtain a combined cross-correlation function, applying an outlier detection or outlier removal approach to identify or remove those segments which are disturbed or corrupted, and re-combining said partial cross-correlation functions without the ones which have been identified as disturbed or corrupted to obtain a less disturbed or less corrupted final cross-correlation function.

A method and apparatus receiving number and using instruction to create resulting number approximating one of square root, reciprocal, or reciprocal square root of number. The resulting number as a product of that process. Using resulting number in a graphics accelerator. Apparatus preferably includes log-calculator, log-arithmetic-unit, and exponential-calculator. At least one of log-calculator and exponential-calculator include implementation non-linear calculator. The non-linear calculators may use at least one of mid-band-filter, outlier-removal-circuit. The invention includes making arithmetic circuit, log-calculator, log-arithmetic-unit and exponential-calculator. The arithmetic circuit, log-calculator, log-arithmetic-unit and exponential-calculator as products of manufacture. The arithmetic circuit may further include at least one of a floating-point-to-log-converter and / or a second of log-calculators. The arithmetic circuit may further include third and fourth log-arithmetic-units providing altered log domain numbers to third and fourth exponential calculators.

A golf ball launch monitor is disclosed that may be used with an alignment stick. The monitor has a default alignment and an image sensor adapted to capture an image of the alignment stick and communicate that image to a processor. The processor is configured to perform the following steps: (a) detect a horizontal edge within the image representative of the alignment stick by detecting large contrast changes; (b) convert each edge to a vector that starts at the sensor's focal point and projects into space based on the sensor's calibration; (c) locate the plane formed by the vectors by applying standard outlier removal and best fit analysis (d) determine the intersection of the plane and an earth tangential plane; and (e) calculate an azimuth alignment angle offset based on the line and the monitor's default alignment. The calculated azimuth alignment angle can then be used to adjust ball flight trajectory calculations.

The invention relates to a method for measuring molecular bitterness of a medicine on the basis of bitterness threshold (BT) concentration. The method comprises the following steps: preparing a seriesof low-concentration solutions of a standard bitter substance and a to-be-tested monomer; at 37 DEG C, performing typical people taste experiment on the series of low-concentration solutions of the standard bitter substance and the to-be-tested monomer by a 'typical psychophysics minimum limit method'; measuring the number of people tasting bitterness under different concentration, performing outlier removal and calculating the final ratio of people tasting bitterness; and establishing a Wible equation between the obtained people number ratio and different concentration. According to the method, the bitterness threshold concentration is established by utilizing the typical people taste experiment and the typical psychophysics minimum limit method, quantitative index for bitterness evaluation of the medicine is provided, technical support is provided for medicine bitter taste masking, the commercial value of the medicine is exploited, the taking effect of the medicine is improved and remarkable economic and social benefits are achieved.

The embodiment of the invention provides a point cloud data processing method and device and an obstacle detection method and device, and belongs to the technical field of computers. The point cloud data processing method comprises the steps of performing three-dimensional reconstruction on a depth image of a current environment to generate a point cloud data set; dividing the point cloud data setinto a plurality of sub-point cloud data sets; performing point cloud filtering processing on each sub-point cloud data set in the plurality of sub-point cloud data sets in a radius outlier removal mode based on a preset standard outlier removal radius and a standard depth interval length; and performing segmentation processing on each sub-point cloud data set after the point cloud filtering processing is performed to obtain a plurality of clustered point cloud clusters. According to the point cloud data processing method, the point cloud filtering effect based on radius outlier removal can be improved, so that the segmentation processing is more convenient and faster.

The invention discloses a Faster-RCNN (Recurrent Convolutional Neural Network) and Kalman filtering combined moving human body tracking method. The method comprises the following steps: firstly, simplifying the Faster-RCNN algorithm, leaving a convolutional neural network and an RPN network, and outputting a moving human body candidate position of the input image through the two networks; then, improving a classic Kalman filtering algorithm, changing a noise covariance matrix defined as a constant matrix in an original algorithm into a time-varying matrix, expanding an original state vector from four dimensions to eight dimensions, and increasing width, height and width and height change rate information of a moving human body position frame in the state vector; and finally, taking the obtained candidate positions of the moving human body as observed values of a Kalman filtering algorithm, obtaining estimated values of a plurality of positions of the moving human body in combination with predicted values of the Kalman filtering algorithm, and solving an average value through least square fitting, outlier removal and residual position removal to obtain optimal estimation of the positions of the moving human body. According to the method, the effect of accurately tracking the moving human body under the dynamic background is achieved.

The invention provides a parallelization method for extracting the melody of counterpoint based on GPU (Graphic Processing Unit), and the method comprises the following three steps: firstly, conducting spectrum transform and pitch value saliency computing to the music data on the GPU; secondly, constructing pitch contour by using pitch saliency on the GPU, getting the relevant characteristics of the pitch contour, and conducting voice detection by using the characteristics; thirdly, conducting frequency doubling removing and outliers removing to the remained pitch contour and finally obtaining the melody track of the counterpoint. The method provided by the invention is based on counterpoint, and can be applied to music without background music and speech sounds. According to the method provided by the invention, the melody can be extracted based on GPU, the extracting time is decreased from the second level to the millisecond level, so as to achieve the standard of real time application. Furthermore, the extracting of required hardware resource is greatly reduced, and the developing speed of the algorithmic is greatly improved, so as to provide the method with wide use value and application prospect in the field of commercial application and scientific research.

A method and apparatus receiving number and using instruction to create resulting number approximating one of square root, reciprocal, or reciprocal square root of number. The resulting number as a product of that process. Using resulting number in a graphics accelerator. Apparatus preferably includes log-calculator, log-arithmetic-unit, and exponential-calculator. At least one of log-calculator and exponential-calculator include implementation non-linear calculator. The non-linear calculators may use at least one of mid-band-filter, outlier-removal-circuit. The invention includes making arithmetic circuit, log-calculator, log-arithmetic-unit and exponential-calculator. The arithmetic circuit, log-calculator, log-arithmetic-unit and exponential-calculator as products of manufacture. The arithmetic circuit may further include at least one of a floating-point-to-log-converter and / or a second of log-calculators. The arithmetic circuit may further include third and fourth log-arithmetic-units providing altered log domain numbers to third and fourth exponential calculators.

Methods are disclosed for removal of outlier pixels from a transmission electron microscopy camera image. One exemplary method includes establishing a desired exposure of n electrons per pixel; exposing the camera to a series of sub-frame exposures to produce a series of sub-frame images; calculating an average image signal of all sub-frame exposures in said series; establishing a threshold selected to achieve a desired number of false positives; evaluating each of said sub-frame exposures for pixels further away from said average than said threshold; and replacing pixels in each of said sub-frame images that exceed said threshold with said average to form corrected sub-frame images.

The invention provides a position control method based on fiber-optic gyroscope integral means. Firstly, output data of M periods of a fiber-optic gyroscope are collected consecutively, outliers are removed from the output data, zero-bias stability Bs of the fiber-optic gyroscope is calculated according to n output data W1', W2'...Wn' with outliers removed. Secondly, smoothing filtering is carried out on the data of the fiber-optic gyroscope after outlier removal. At last, a transposition angle value of a servo-stabilization platform is obtained by integrating the angular rate wp of the fiber-optic gyroscope after smoothing filtering is carried out. The position control method based on the fiber-optic gyroscope integral means can be widely applied to a position twirl control system of a milliradian grade, for example a photoelectric instrument indicated by a helmet and a camera system indicated by a radar. Turning precision of 1 mrad can be qualified for introducing an objective to fields of view through position twirl. If no target indication high precision is required, optoelectronic equipment of this kind can entirely adopts the technology, and utilization of a light code disc or a resolver can be omitted; equipment cost can be saved.

The invention relates to an ADCP-based seawater flow velocity data processing method. The method comprises the following steps: setting and putting an ADCP instrument; carrying out data extraction andconversion; carrying out data merging and grouping; to be specific, dividing all the converted data into a plurality of '. Mat' files numbered according to a time sequence, each file containing measurement results of all variables in a certain period of time, and needing data merging; complementing the data; and removing a data abnormal value.

Disclosed are a homography estimation apparatus and method. A down-sampling unit reduces the reference image and the target image to the same size to generate down-sampled images. A feature point extraction unit extracts feature points from the reference image and the target image down-sampled to the same size, respectively. An outlier removal unit matches the feature points extracted from the reference image with the feature points extracted from the target image, and detects and removes outliers, which are feature points that do not match, from the matched feature points. A homography estimation unit estimates a homography using feature points from which outliers are removed. A reference image correction unit corrects the reference image using the estimated homography.

The invention discloses a real-time video splicing method based on multiple unmanned aerial vehicles. The method comprises the following steps: step 1, correcting postures of the unmanned aerial vehicles; performing adjusting in the global coordinate system by using the mutual mapping relation of the image pixel coordinate system and the global coordinate system; step 2, carrying out image shooting; step 3, carrying ot image processing and feature extraction; carrying out image processing and feature extraction by utilizing an unmanned aerial vehicle airborne CPU; 4, enabling the ground station to receive the processed image information; 5, searching a feature matching area; step 6, deleting abnormal values; deleting the abnormal value by adopting an improved random sampling consistency algorithm; step 7, optimizing video splicing quality; and step 8, splicing video output. According to the invention, the workload of a ground station is reduced by adopting a distributed feature extraction technology, the splicing efficiency is improved by utilizing the prompt of a flight controller, the splicing speed of images shot by a plurality of unmanned aerial vehicles is improved, the splicing quality is improved, and the real-time splicing output of the images shot by the plurality of unmanned aerial vehicles is realized.

The invention relates to the technical field of computer vision application, in particular to a building monomer instance segmentation method and device, and the method comprises the steps: carrying out the noise reduction of a point cloud generated by an unmanned plane image, and removing outliers; overturning the image point cloud after noise reduction and outlier removal, and removing the ground; semantic segmentation is carried out on the image, and points of which the categories are buildings are screened out from the point cloud of the whole image; performing spatial density clustering on the points of the building to obtain a point cloud cluster family forming each building monomer; processing each point cloud cluster family to generate a three-dimensional frame of the family; and carrying out frame cutting and secondary semantic segmentation on the three-dimensional frame to obtain a complete building point cloud monomer. The method and the device support single segmentation of point clouds generated at lower cost, and have strong robustness. The modeling stability is greatly improved by using the ground filtering algorithm, the calculation complexity is greatly reduced, the operation speed is accelerated, and the model is more accurate by using the deep learning method.