67 results about "Statistical filtering" patented technology

Systems and methods for the design or implementation of statistical filters for use in the spectral shaping of transmissions are disclosed. A desired power spectrum may be mapped to find pole locations that approximate the desired spectrum. These pole locations may then be mapped to the edge or inside of an equilateral polygon lying inside a unit circle, the equilateral polygon having the same number of sides as the order of the statistical filter desired and one vertex mapped to unity, to yield a set of eigenvalues. These eigenvalues may be the eigenvalues of a stochastic matrix the elements of which may be the Markov transition probabilities for use in a statistical filter designed to achieve the desired power spectrum. Use of a statistical filter employing these Markov transition values may be utilized to shape UWB or other signals to achieve the desired power spectrum.

A system for conducting transistor performance analysis is disclosed, in which automatic graph plotting is interactively enhanced with human judgment. The system includes executing transistor performance analysis software, which receives templates specifying parameters, graph charting options, and algorithms, as well as a database of transistor values, as its inputs. The software produces an output document with linked graphs and a summary report. The software extracts, filters and applies statistical regression to large quantities of data. The software also applies statistical filtering to the data and automatically plots hundreds of charts and graphs based on the data. Graphs are color-coded to highlight relationships that suffer from unusually high noise in the data. Users can manually adjust lines on the graphs, which are automatically reflected in dependent graphs and the summary report. Changes to program methodology can be achieved by changing the template rather than by modifying the software.

This invention generally discusses wireless devices, servers and communications networks. In particular the invention pertains to performing observations in one or more mobile terminals and processing and distributing the related data in a server side system through layered data processing activities, and conversion of non- parametric data into parameterized form through the utilization of statistical filtering and semantic data structures. It is further explained how such multi-layer, parametrized data can be utilized for predictive purposes, and how feedback loops can be built with the physical world to improve future predictions. The invention is applicable in various applications, for example in systems where precise digital profiles of users need to be built on a continuous basis, and such profiles need to be dynamically linked to one or several actions triggered by emerging characteristics in the data. The multi-layer approaches makes it possible to structure output statistics into continuous and standardized, periodic datasets, even though input data is highly unorganized, non-chronological, and sporadic. Similarly, the invention describes how the multi-layer data storage structure and chosen statistical operations make it possible to build virtually an infinite number of further aggregations and averages based on the output data streams.

The invention provides a rapid analysis and processing method for infrared and Raman spectrum data on the basis of artificial intelligence. The rapid analysis and processing method comprises the following steps of: acquiring an infrared and Raman spectrum library of a single-sample learning group; carrying out statistical filtering on data of the spectrum library to remove suspect and invalid data; carrying out unsupervised learning classification on the spectrum library of the learning group by artificial intelligence; carrying out chemical analysis, verification and classification on every type of samples distinguished by classification results; if a sampling result can not meet the application needs, returning an artificial intelligence program and carrying out supervised learning classification by using chemical analysis results; by multiple repeated classification and iteration, using the artificial intelligence program to generate an analysis prediction model capable of meeting the application needs automatically; and using the model for rapid spectrum analysis and detection of unknown samples in the type of samples. The rapid analysis and processing method provided by the invention is applicable to rapid spectrum analysis and detection of a large amount of samples, can realize accuracy adjustment according to the needs, and can be rapidly applied in different sample testing scenes, so that the application prospect and potential are great.

The invention provides a business data multidimensional statistics method and system. The method comprises the steps of displaying dimensions on a multidimensional statistics page according to dimension names configured in a background; writing information of all business data in a business data total table, extracting the data in the business data total table, and generating four simplified tables according to interfaces, the dimensions, servers and sites; from the business data total table, extracting aggregation information of dimension values of the dimensions of the business data according to different time granularities; according to clicking operation of a user, refreshing a list of the dimensions capable of executing drill-down operation in current operation dimensions; or, according to the clicking operation of the user, generating a time period statistical range, a dimension value statistical filtering condition and dimensions needed to be subjected to statistics. The multidimensional statistics configuration is simplified; different layers of drilling can be displayed on the same page at the same time; re-query is not required during drill-up; and the dimensional data statistics page displays a result quickly.

The invention relates to the field of mobile laser radar data processing, and discloses a universal automatic extraction method for a mobile laser radar data road surface, which comprises the following steps: step 1, eliminating outliers by using a statistical filtering method, and extracting road surface candidate points from laser radar data according to trajectory data and elevation information; step 2, segmenting the candidate points into a plurality of pseudo scanning lines along the advancing direction of the vehicle, and converting a coordinate system of the point cloud into a local coordinate system based on the pseudo scanning lines; and step 3, achieving road surface extraction by detecting height mutation between the road surface and the kerb or the drainage ditch. The inventionfurther discloses an automatic extraction method of the road surface marker line. According to the method, the trajectory data, the three-dimensional coordinates of the point cloud, the intensity information and the intensity gradient information are comprehensively utilized, different types of mobile laser radar data are effectively organized, the road surface and the marker line are automatically extracted, and the method is generally suitable for different types of mobile laser radar systems.

A method for real-time restoration of a fog-degraded image with white balance correction relates to the image information processing field and solves the problems of complex algorithm, image shift in image processing process and unachievable embedded platforms such as DSP (Digital Signal Processor) and the like in the existing fog image restoration method. The method comprises the following steps of: performing normalization operation on an original fog-degraded image in the atmosphere background and extracting the minimum channel image of the original fog-degraded image, and obtaining a dark channel image by a statistical filtering quick method with O (N) complexity; then, obtaining the transmission parameter image of the original fog-degraded image; obtaining the restored image of the normalized fog-degraded image by using the transmission parameter image according to an atmosphere optical imaging model; and finally, according to the white balance theory, mapping the normalized restored image range to [0, 255] by taking the atmosphere background light as pure white. As a result, the output image obtained through white balance global mapping is the final restored image. The method provided by the invention is capable of meeting various outdoor scene monitoring application requirements.

A method for automatically constructing a planogram from photographs of shelving, replacing laborious manual construction includes the following steps: a step (1) in which the images are received, a step (2) in which the images are assembled, a step (3) in which the structure is automatically constructed, a step (4) in which the products are automatically detected, and a step (5) in which the products are positioned in the structure. The product detection step (4) enhances traditional image recognition techniques, using artificial learning techniques to incorporate characteristics specific to the planograms. This product detection step (4) also includes at least two successive classification steps, namely: an initialization step (41) with detection of product categories; and a classification step (42) with the classification of the products themselves, each of these steps including a first image recognition step, followed by a statistical filtering step based on the characteristics specific to the planograms.

The present invention provides systems and methods that employ statistical filtering and ranking techniques to improve content search engine search results by tuning a general-purpose search engine for an entry point for a group of users. The filter can be manually and/or automatically configured via providing training sets of relevant and non-relevant data. For example, a relevant set of data comprising web pages associated with a desired search context for the group utilizing the entry point can be employed. In addition, a non-relevant set of data comprising random and unrelated documents can be employed. The learned filter can then be employed to compare a returned result with the known relevant and the known non-relevant data sets to determine whether the result is relevant to the user for the entry point, and the degree of relevance. The results can then be presented to the user based on the relevance.

The invention relates to a statistical filtering infrared small target detection tracking method based on YOLOv5 and Deepsort, and the method comprises the steps: S1, collecting an infrared small target image under a complex background through an infrared imaging device, and obtaining an infrared small target image data set; s2, preprocessing the infrared small target image data set to obtain a preprocessed image data set, and dividing the preprocessed image data set into a training set and a verification set; s3, a YOLOv5s model in the YOLOv5 algorithm is trained, and a training set of a Deepsort model is obtained; step S4, inputting the training set of the Deepsort model into the Deepsort model for training, and constructing an infrared small target detection and tracking recognizer; and S5, performing real-time detection and tracking on the infrared small target by using the infrared small target detection and tracking identifier. According to the method, the infrared small target under the complex background can be accurately and quickly detected, and the robustness and the detection rate are improved. In addition, the real-time tracking effect can be achieved.

The invention relates to a ship over-limit early warning method and system based on point cloud data. The method comprises the following steps: acquiring environmental point cloud data acquired by a laser radar; preprocessing the environmental point cloud data through a statistical filtering algorithm and a bilateral filtering algorithm; performing two-dimensional gridding on the preprocessed environment point cloud data, and performing segmentation according to a K-means clustering algorithm to obtain target ship point cloud data; obtaining a height value corresponding to the highest point of the point cloud data of the target ship; and comparing the height value with a preset height limit value, and when the height value is larger than or equal to the height limit value, sending out an over-limit early warning signal of the corresponding ship. According to the technical scheme of the invention, the early-warning monitoring precision is ensured by the early-stage multi-dimensional preprocessing algorithm, and the processing efficiency of the early-warning monitoring process is ensured by the later-stage two-dimensional gridding processing, so that efficient and accurate early warning of the passing ships is realized.

The invention discloses a plant point cloud collection registration and optimization method based on a TOF camera. The plant point cloud collection registration and optimization method comprises the following steps: (1) placing a detected plant on a turntable, adjusting the visual angle of the TOF camera, and enabling the TOF camera to be aligned with the detected plant; (2) starting a rotary table to enable the rotary table to rotate automatically, and collecting point cloud data of a measured plant at intervals by a TOF camera; (3) performing real-time preprocessing on each frame of point cloud data acquired by the TOF camera; (4) performing registration and optimization processing on each two adjacent frames of point cloud data in sequence, and then performing integration to obtain complete plant point cloud data; and (5) eliminating discrete noise in the plant point cloud number obtained by registration and optimization processing by adopting statistical filtering processing to obtain final point cloud data. The plant point cloud collection registration and optimization method provided by the invention can effectively solve the local point cloud layering phenomenon.

The invention provides a two-stage multi-mode three-dimensional instance segmentation method. The method comprises two parts of two-dimensional prior information acquisition and three-dimensional instance segmentation. The two-dimensional prior information acquisition part adopts an RGBD multi-mode fusion two-dimensional instance segmentation network, a depth feature correction module and a data enhancement strategy for a depth map are designed for the problem of low quality of the depth map, and a feature alignment module is added in a feature pyramid module. And the three-dimensional instance segmentation part is used for generating a point cloud with instance segmentation information by using a coordinate transformation method in combination with a depth map according to the acquired two-dimensional prior information, and then removing noise points in the point cloud by using direct filtering and statistical filtering to realize three-dimensional instance segmentation of a target. The two-stage multi-modal three-dimensional instance segmentation method has the beneficial effects that the two-stage multi-modal three-dimensional instance segmentation method can perform accurate three-dimensional instance segmentation on a scene target in an application scene with a small amount of two-dimensional annotation data and without three-dimensional annotation data, and has the advantage of high real-time performance.

The invention discloses a pig body size parameter measurement method based on point cloud, and relates to the field of whole pig measurement. According to the method, a Kinect calibration method is used for calibrating Kinect; related parameters of a Kinect camera of the system are obtained; two Kinects are erected at the top of a pig house; the distance between the two cameras is set to be 3000 mm to 4000 mm. The cameras are connected with a computer through a connecting device; a hardware system for point cloud detection is formed; pig point cloud data is collected; the acquired point clouddata is denoised through conditional filtering, statistical filtering and voxel filtering denoising algorithms to remove outliers through multi-element Gaussian distribution abnormal point detection,point cloud coarse registration is carried out on denoised point cloud through SAC-IA, point cloud fine registration is made through an ICP algorithm to obtain pig overall point cloud, and pig body size parameters are acquired through a geometrical characteristic extraction algorithm. The method has the advantages of high sensitivity, high precision and low labor intensity, and has very importantsignificance.

The invention relates to a distance and course measurement method based on a laser radar. According to the method, on the basis of 16-line laser radar, a distance between wall surfaces on the two sides of the robot in a long corridor environment and a yaw angle of the robot in the advancing direction are measured. The method comprises the following steps of 1) installing a laser radar sensor on a rotary holder of a robot, and scanning to obtain three-dimensional point cloud data of an environment; 2) extracting features required by measurement from the scanned three-dimensional point cloud data, wherein the specific operation comprises steps of performing straight-through filtering, statistical filtering, European clustering and random sampling consistency segmentation on the obtained three-dimensional point cloud, and calculating the centroid of the clustered point cloud cluster; and 3) according to the obtained centroid and plane parameters of the point cloud cluster, calculating a distance from the laser radar to the wall surfaces on the two sides of the corridor and the yaw angle of the robot. According to the method, requirements of the robot on the distance to the wall surfaces on the two sides and the yaw angle in autonomous navigation in the long corridor environment are met, and autonomous navigation of the robot in a specific scene is realized through provided data.

The invention discloses a laser three-dimensional point cloud and CT three-dimensional point cloud registration method, and the method comprises the steps: carrying out the statistical filtering, point cloud region growth processing and voxel filtering of the body surface laser three-dimensional point cloud data of a patient, and reconstructing the 4D-CT data of the patient into CT three-dimensional point cloud data; selecting a point set S1 from the laser three-dimensional point cloud data, and selecting a point set S2 from the CT three-dimensional point cloud data; extracting a feature pointset Q from S1 and extracting a feature point set V from S2 by using a Point RegNet network; setting all the feature points in Q to be connected with one another to form a set ES of feature edges, setting all the feature points in V to be connected with the feature points adjacent to K to form a set E of feature edges, and registering the graphs (V, E) with the graphs (Q, ES); distributing weightsto the points in the S1 by utilizing the motion amplitude information, establishing a GMM model on the basis of the S2, and constructing a probability likelihood function by combining the GMM model with the points in the S1 after the weights are distributed. According to the technical scheme, the problem that the prior art cannot be directly applied to body surface point cloud registration is solved.

A method for removing defective pixel image-artifacts includes an adaptive reconstruction kernel taking up to four main and diagonal, defect-free sub-kernel directions into account composed for each defective image pixel. The defective pixels impacted image is real-time corrected by statistical filtering or by a weighed directional convolution of kernel-associated replacement values, calculated by means of an advanced multi-parabolic reconstruction algorithm, for each contributing sub-kernel direction based on 5×5 pixels neighborhood image data readily accessible via a predetermined AMP kernels image-offsets structure.

The invention relates to an unmarked planar flexible body deformation tracking method based on NURBS. The method comprises the following steps that 1, after an original point cloud containing a planarflexible object point cloud is obtained through a Kinect depth camera, a point cloud of a target planar flexible body needs to be segmented through a certain process, and a noise point cloud in the space is removed through radius filtering or statistical filtering, so that a pure target point cloud is obtained; 2, the obtained target point cloud is initialized to obtain an initial NURBS curved surface; 3, an approximate NURBS curved surface is generated through curved surface fitting, the NURBS curved surface is composed of tensor products of the two NURBS curved surfaces, and the NURBS curved surface fitting result is obtained through a curved surface fitting method for the two NURBS curved surfaces; and 4, the curved surface fitting of the unmarked surface-mounted flexible body is completed, and the variation of a certain point on the rear surface-mounted flexible body before and after deformation is obtained through an NURBS Debull algorithm.

A method for reducing image disturbances caused by reconstructed defective pixel clusters located in signal-gradient affected diagnostic image regions. An individually adapted central symmetrical pair reconstruction (CSP) kernel is composed for a defective image pixel based on a kernel-pair candidate order encoded in a model thereby using the pixel's validity state. The image impacted by defective pixels is corrected in real-time by statistical filtering or spatial convolution of the kernel-associated image data accessible via a predetermined CSP kernels image-offsets structure.

The invention discloses a blade profile extraction method based on blade profile measurement point cloud. Firstly, the invention studies a preprocessing method for improving profile reconstruction quality and speed, and the preprocessing method comprises a plurality of steps of segmentation, statistical filtering and simplification based on an octree structure. Secondly, the method includes researching a profile reconstruction method of a grid envelope, and performing spatial interpolation on a fixed-height blade profile to calculate accurate blade profile data; specifically, the method adoptsan outlier filtering method based on distance statistics to carry out data filtering, and includes preprocessing the point cloud data based on octree data structure point cloud simplification and other technologies; then, combining reconstruction algorithm based on region growth with a Delaunay criterion to realize rapid reconstruction of the blade point cloud; and finally, intersecting a plane with a specified height in a part drawing with a topological grid, and finishing extraction of blade profile data through spatial interpolation. The method is high in calculation efficiency, overcomesthe problem that the blade profile characteristics are difficult to accurately reserve in a projection method, and has a good engineering application prospect.

The invention belongs to the technical field of electronic reconnaissance, and particularly relates to a passive radar signal detection method based on sequential statistical filtering and binary detection. According to the invention, after certain speed reduction and signal-to-noise ratio improvement are carried out on low-signal-to-noise-ratio signals in a large instantaneous monitoring bandwidth through channelization preprocessing, binary accumulation detection is combined, and rapid and accurate detection can be realized under a certain low-signal-to-noise-ratio condition. The method adapts to the trend that the sampling rate of a reconnaissance system in the passive radar field is higher and higher, and solves the problem that signal detection is difficult under the condition of low signal-to-noise ratio. On the basis of a relatively stable condition in a channel, a detection threshold is estimated by performing sequential statistical filtering processing on reference data in a parallel sliding window without additional threshold compensation; wherein the parallel pipeline type structure ensures that an adaptive constant false alarm threshold can be quickly obtained under strong noise, the detection and false alarm probability can be further improved by adopting a binary accumulation detection technology, so that the application limitation of an existing detection method is overcome.

Image alias rejection when converting a high resolution rasterized waveform to a lower resolution rasterized waveform for display uses a statistical filter. The statistical filter provides a shaped probability density function either by combining the outputs of multiple random number generators, such as linear feedback shift registers, or by using a corresponding look-up table to produce a dither signal. The statistical filter may be applied to one or both of the dimensional values for each data point of the high resolution rasterized waveform by combining the dimensional values with the dither signal. The resulting filtered dimensional values may then be subsampled, such as by truncation, to produce values for a lower resolution rasterized waveform display.

The invention discloses a statistical and bilateral filtering point cloud denoising method based on improved neighborhood search, and the method comprises the following steps: 1, carrying out the processing of point cloud data through employing a statistical filtering algorithm based on improved neighborhood search, and obtaining the processed point cloud data: 1.1, obtaining a point cloud two-dimensional plane projection drawing; 1.2, carrying out grid division to obtain a plurality of rectangles; 1.3, primarily removing outliers; 1.4, calculating an average distance corresponding to any point p; 1.5, calculating an average distance mu and a standard deviation sigma; 1.6, calculating a maximum threshold value dmax, and removing outlier noise points; and 2, performing bilateral filtering on the processed point cloud data obtained in the step 1 to obtain filtered point cloud data. According to the method, the calculation amount is obviously reduced, and the search speed is increased; point cloud data are processed in combination with a bilateral filtering algorithm, large-scale outlier noise points are removed, small-scale noise points in the point cloud are also removed, and the limitation of independent use of the two algorithms is solved.

The invention discloses a blade machining cutter compensation amount calculation method, which comprises the following steps of: selecting a group of measurement points on a blade section line, and acquiring actual position data of each measurement point; fitting to obtain the location degree of the cross section corresponding to the cross section line; judging whether the location degree meets a tolerance requirement or not; calculating the error of each measuring point to obtain a group of error data, and calculating the kurtosis of the error data; performing statistical filtering on the error data to enable the kurtosis of the error data after statistical filtering to meet the requirement; judging whether the blade profile contour is qualified or not; calculating the variance of the error data meeting the kurtosis requirement; judging whether the variance meets a variance threshold value or not; and the cutter compensation amount is calculated according to the compensation coefficient and the mean value of the error data meeting the kurtosis requirement. According to the method, the blade machining state and the cutter compensation amount can be accurately obtained in the blade finish machining stage, and the production auxiliary time in the blade machining process can be remarkably shortened while the machining quality is effectively guaranteed.