185 results about "Dbscan clustering" patented technology

The invention discloses a method and device for determining a taxi pick-up and drop-off hot spot region center. The method includes the steps of obtaining taxi state information data, extracting geographical location data according to taxi serial numbers, ranking the geographical location data according to timestamps, and obtaining taxi track data; extracting pick-up and drop-off position data from the taxi track data, wherein the datum of each pick-up spot or drop-off spot of each taxi is a sample datum; carrying out DBSCAN clustering on the pick-up and drop-off position data primarily, and extracting out core object data, wherein the number of the sample data within the preset scanning radius of a core object is greater than or equal to a first preset threshold value; merging the core object data within the selected time period, carrying out secondary DBSCAN clustering on the merged core object data, extracting the cluster core object data, and determining the taxi pick-up and drop-off hot spot region center on the basis of the cluster core object data.

The invention discloses a multi-target tracking method based on millimeter-wave radar. The method comprises acquiring point cloud data by millimeter-wave radar; clustering the point cloud data to distinguish the echo signals of different targets; and estimating the state information of multiple observed targets according to a clustering result so as to track the multiple targets. The method improves a DBSCAN clustering algorithm according to the characteristics of the millimeter-wave radar, improves the accuracy of esimating the number of the targets and the states of the targets, and predictsand tracks the trajectories of the multiple targets in a complex environment by using a Kalman filter and data association algorithm.

The invention provides an acquisition method and apparatus of the operation area of farm machinery. The acquisition method of the operation area of farm machinery includes the steps: receiving the farm machinery operation track data sent from a farm machinery positioning device; based on the operation speed, improving the neighborhood radius determination method of the dbscan clustering algorithm; utilizing the improved dbscan clustering algorithm to filter the road driving point and the field transition point in the farm machinery operation track data; according to the filtered farm machinery operation track data, determining the number of fields operated by the farm machinery; and utilizing the distance method to calculate the area of each field operated by the farm machinery. The acquisition method and apparatus of the operation area of farm machinery can adaptively determine the neighborhood radius according to the operation speed of farm machinery, can effectively segment the adjacent fields operated by the farm machinery, and can improve the accuracy and automation degree for acquisition of the operation area of the farm machinery.

The invention provides an airborne laser point cloud classification method based on a high-order conditional random field. The airborne laser point cloud classification method specifically comprises the following steps: (1) point cloud segmentation based on DBSCAN clustering; (2) point cloud over-segmentation based on the K-means cluster; (3) construction of a point set adjacency relation based onthe Meanshift clustering; and (4) construction of a point cloud classification method of a high-order conditional random field based on the multi-level point set. The method has the advantages that:(1) a multi-layer clustering point set structure construction method is provided, and a connection relation between point sets is constructed by introducing a Meanshift point set cluster constrained by category labels, so that the categories of the point sets can be classified more accurately; (2) a multi-level point set of the non-linear point cloud number can be adaptively constructed, and information such as the structure and the shape of a point cloud target can be more completely represented; and (3) a CRF model is constructed by taking the point set as a first-order item, and higher efficiency and a classification effect are achieved, so that a higher framework is integrated, and a better effect is obtained.

The invention discloses a method for detecting a full line of a lane based on optical flow point locus statistics. The method comprises the steps of step 1, mounting a camera above the middle part of a one-way road; step 2, performing preprocessing on a video frame image; step 3, acquiring optical flow point sets of moving vehicles; step 4, dividing the optical flow point set of each moving vehicle by means of a DBSCAN clustering algorithm, and representing the divided optical flow point sets by a rectangular area with a fixed size; step 5, performing statistical stacking on the divided area of each moving vehicle, and performing binarization processing on a result; step 6, performing linear fitting on the middle point set of contour points of a white pixel area which accords with a condition in a binary image; and step 7, determining the full line of the lane by means of a line which is obtained through fitting the middle point set. The method provided by the invention is not affected by illumination, weather, vehicle and road condition. Furthermore the method realizes high detection precision for the full line of the lane and high robustness.

The invention provides a method for extracting a target classic track in a complex environment. The method provided by the invention can realize extraction of the target classic track under complex conditions such as strong data randomness, loud noise, different track lengths and large point track interval difference of the target track, and is realized by the following technical scheme: reading target track data, prescreening according to value range constraints with different attributes in the track data, and removing abnormal data items; rasterizing latitude-longitude grids, counting the point track density in each latitude-longitude grid, and extracting all the track data and filtering the track data with relatively high point track density; clustering the filtered track data with a multi-distance threshold DBSCAN clustering algorithm to form a plurality of track clusters; and setting a scanning line for each cluster, sequentially calculating an average value of intersection pointsof the scanning lines and the track to obtain the point tracks in the target classic track, and connecting the point tracks in order to obtain a target classic track extraction result.

The invention provides a driver fixation point clustering method based on a density clustering method and a morphology clustering method, and belongs to the field of typical density clustering methods and mathematic morphology clustering methods. The driver fixation point clustering method includes the steps of putting forward a density method and mathematic morphology method combined self-adaption DBSCAN-MMC method, applying the method to driver fixation point clustering, setting the value of the Eps through fixation point structure parameters, obtaining an initial point set of MMC clusters through the DBSCAN, determining the number of the clusters, reducing outliers produced through DBSCAN clusters through the self-adaption MMC clusters, and completing clustering oriented to driver fixation areas. According to the method, the advantages of irregular shape clustering of the DBSCAN and the MMC are fully used, the defects of the two clustering methods are overcome, the clustering effect is superior to the clustering effect of the conventional DBSCAN clustering method and the conventional MMC clustering method when the driver fixation areas are divided, and the driver fixation clustering quality is improved.

The present invention relates to a crowd grouping evacuation simulation method and system based on a half DBSCAN clustering algorithm. The method comprises: creating an evacuation scene model and a crowd individual model, and initializing evacuation scene information and crowd individual information, so as to obtain initialization parameters; according to the initialization parameters, dividing a crowd into multiple groups by using a half DBSCAN algorithm, so as to obtain a crowd grouping result; and using the crowd grouping result as an initial state of crowd motion, and making a local path plan by using a social force model, so as to obtain a motion path of each individual in an evacuation scene and perform crowd evacuation simulation. According to the method, the half DBSCAN clustering algorithm is combined with the social force model to complete crowd evacuation simulation in a complex scene, so that crowd evacuation simulation in the complex scene is really and effectively realized, the accuracy of crowd evacuation simulation is improved, and the grouping phenomenon in the evacuation process is demonstrated vividly.

The invention discloses a mobile phone signaling data-based urban transportation characteristic determining method, which comprises the steps of preprocessing mobile phone signaling data and quickly extracting an available field to form a storage format employing a user ID as a key field; sorting mobile positions of each user according to time and carrying out denoising according to a speed and angle exception judgment method; forming a convergence point by using a DBSCAN clustering algorithm, and identifying all stay points and stay time of each user every time; and classifying the stay points and calculating the urban traffic generation amount, the traffic attraction amount, the travel times, the average travel times, the travel distance, the commuting trip distance, the living index and the employment index. The urban traffic generation amount, the traffic attraction amount, the travel times, the average travel times, the travel distance, the commuting trip distance, the living index and the employment index are quickly and relatively accurately calculated by fully utilizing mobile phone signaling data, and data support is provided for urban transportation planning, traffic management and traffic strategy research.

The present invention discloses a wind turbine generator operation state identification method based on a DBSCAN. The method comprises the following steps: the step 1: collecting operation data of a wind turbine generator and basic parameters of the wind turbine generator; the step 2: performing data normalization processing of the wind turbine generator; the step 3: screening the shutdown data from the operation data according to the wind speed and the power data; the step 4: taking the wind speed, the power and the pitch angle data obtained through the normalization processing in the step 2 as the input data of the DBSCAN, and performing optimization of the model parameters of the DBSCAN clustering; and the step 5: obtaining the operation state identification result of the wind turbine generator according to the data classification results of the step 3 and the step 4.

The invention provides an unstructured road detection method based on four-line laser radar. The method comprises the steps that three-dimensional information coordinate points of a road are acquired by the four-line laser radar on a vehicle and transformed to the coordinate points under the coordinates at which the vehicle is located; the coordinate points are preprocessed to eliminate abnormal data and noise interference; corrosion and expansion processing is performed on the preprocessed data, and clustering of the result after corrosion and expansion is performed by using a DBSCAN clustering algorithm in a vehicle body Z-direction and ground points are extracted, wherein the Z-axis positive direction is upward in a way of being perpendicular to the horizontal ground; straight line fitting is performed on the ground points by using hall transformation so as to obtain the straight line scanned on the ground by the laser radar; the points of the two ends of the straight line are taken as the road boundary points of the scanning line, and smooth tracking of the road boundary points is performed by using Kalman filtering so as to obtain the final road boundary points; and road boundary fitting is performed on the final road boundary points by using the least square method.

The invention provides a pedestrian re-identification method and device based on a semi-supervised training mode and a storage medium. The method comprises the following steps: constructing a trainingsample set by using the collected target domain sample set and open source data set based on a semi-supervised training mode; and training a deep residual network by using the training sample set toobtain a trained deep residual network model, identifying the acquired pedestrian images by using the trained deep residual network model to obtain feature values of the pedestrian images, and determining whether the pedestrian images belong to the same person according to cosine distances among the feature values. According to the invention, the virtual sample is generated; a smoothing function is constructed when a virtual sample is generated; and meanwhile, a DBSCAN clustering algorithm is used for adding pseudo tags to the virtual samples, local features and global features are used in thedeep neural network, and a joint loss function of different weight combinations is adopted, so that the trained deep neural network is accurate and more reliable in recognition.

The present invention provides a locomotive and vehicle abnormal axle temperature diagnostic method and system. Temperature time-domain features of a plurality of associated measurement points at a plurality of time windows are combined to a feature set, the feature set is subjected to k-means clustering, the feature position difference is determined through the k-means clustering to determine whether there is an isolated measurement point or not, if yes, the maximum radius of a cluster where normal measurement points belong to is taken as a radius of neighborhood, the number of the temperature time-domain features corresponding to a single associated measurement pint is taken as the minimum neighborhood density, the feature set is subjected to DBSCAN clustering according to the radius ofneighborhood and the minimum neighborhood density to determine whether the distribution density has obvious difference or not. If the results of the k-means clustering and the DBSCAN clustering of thetemperature time-domain features corresponding to a certain associated measurement point have the position difference and distribution density difference at the same time, it is determined that the associated measurement points have temperature anomaly. The method and the system provided by the invention effectively improve the locomotive and vehicle abnormal axle temperature diagnosis accuracy and correspondingly reduce the diagnosis misjudgment rate.

The invention discloses a re-clustering-based pushing method and apparatus, a computer device and a storage medium. This method includes the steps of obtaining clusters for classification by users through DBSCAN clustering on each row vector in a scoring matrix and a sub-user-scoring matrix corresponding to each cluster one by one, obtaining a row vector selected in the scoring matrix as a targetuser, obtaining a similar user group scoring matrix of a target user and a commodity recommendation line vector corresponding to the similar user group scoring matrix according to the sub-user--scoring matrix of the target user, obtaining a commodity recommendation list according to the commodity recommendation line vector, and pushing the commodity recommendation list to a receiving end corresponding to the target user. This method enables the user-scoring matrix to be divided into a plurality of sub-matrices to carry on the maintenance separately, reduces the maintenance cost, and can carryon the commodity information push to the target user accurately according to the sub-matrices horizontally.

The invention discloses a processing method and system for power equipment to monitor abnormal data in an online manner. The method comprises the following implementation steps of carrying out abnormal data positioning analysis on an original data set, and selecting a DBSCAN clustering algorithm to carry out missing value filling to complete data cleaning or an abnormal data cleaning method basedon an association rule to complete data cleaning; calculating each data quality index value in a preset data quality evaluation system, and generating a data quality evaluation report; and displayingand outputting an abnormal data positioning analysis result and a data quality evaluation report. According to the invention, the accurate and effective cleaning of abnormal data can be realized; thecleaning effect is good, the visual display of the data quality before and after data cleaning can be realized, the proportion and the reason of abnormal data generation are statistically displayed, power grid workers can take corresponding measures to improve the data acquisition and uploading process, and the generation of abnormal data is reduced from the source, so that the workload of data processing is reduced, and the working efficiency and the accuracy are improved.

The invention provides a tour route recommendation method combined with short term traffic flow forecasting. The method includes the steps that historical picture description information is downloaded from a picture sharing website, pictures are clustered through the DBScan clustering algorithm to obtain interest points large in picture density, historical tour routes of the historical picture description information are extracted, time sequence modes, at different time intervals, of the interest points are built according to the historical picture description information, a probability transfer matrix of a Markov model is generated according to the historical tour routes in a historical tour route set in a region, the probability transfer matrix of the Markov model is updated according to pictures and the picture description information uploaded by a user in real time, and the interest point maximum in transfer probability is recommended to the user according to the updated probability transfer matrix of the Markov model. The method is combined with short term traffic flow of the interest points for route recommendation, and has the function of dispersing passenger flow in scenic spots.

The present invention discloses a travel and activity mode identification method based on a BDSCAN clustering algorithm. The method comprises the following steps: cleaning traveler spatio-temporal track data sets that are continuously acquired; calculating an average speed of coordinate points of cleaned data sets, and classifying a position coordinate point whose average speed is higher than a set threshold into a travel mode; based on a DBSCAN clustering algorithm, performing clustering analysis on the cleaned data sets, and determining an activity starting point and an activity ending point according to a clustering result; and according to coordinates and time of data points of an identified travel mode and activity mode, generating a travel time table. According to the method disclosed by the present invention, based on the acquired traveler spatio-temporal track sequence sets, the behavior modes of the travelers are divided into the travel mode and the activity mode by using the density-based clustering algorithm (DBSCAN). The method disclosed by the present invention is convenient for calculation and actual operation and has strong practicality, and by the method, the behavior mode of the travelers can be determined more accurately, so as to facilitate subsequent researches, so that the method has important realistic significance.

The invention relates to a layered point cloud segmentation method based on DBSCAN. Firstly, a CSF is adopted to separate a ground point from a non-ground point; In a non-ground point segmentation process, point clouds in the vertical direction are layered according to a certain height, DBSCAN clustering is carried out on projection points of each layer on an XOY plane to obtain a central point ofeach cluster, all the clustered central points are projected to the XOY plane, and each object main body is clustered by utilizing DBSCAN to obtain a plurality of object main bodies; a judgment is made whether each main body point exists in each layer of each main body or not, judging the number of objects contained in each cluster, and finally, a cluster of multiple objects is segmented. For segmentation of side viewpoint cloud data, extraction of most of main bodies in a scene can be guaranteed, certain robustness is achieved, particularly, the invention has good performance in the scene with trees as the main component, and the result obtained through the method has certain significance in point cloud classification and point cloud three-dimensional reconstruction after point cloud segmentation.

The invention discloses a dynamic and static characteristic-based video classification method. The problem that the video classification accuracy is not high is solved. The method comprises the stepsthat dynamic characteristics and static characteristics in video are processed, information is fused by means of Cholesky conversion, and then video classification is completed by using a GRU neural network; the dynamic characteristics of each video frame are captured through a DT algorithm, all video frames are isolated through a DBSCAN clustering algorithm, a motion frame is built in each frameof each video clip, the motion frames between the adjacent frames of each video clip are connected, and capturing and tracking of the dynamic characteristics are completed; by means of HoG and BoW methods, a dynamic information histogram generated by the dynamic characteristics and a static information histogram generated by a CNN neural network are fused by means of Cholesky conversion; finally,the GRU neural network is utilized for achieving video classification. Accordingly, by processing the dynamic and static information separately, the accuracy of video classification can be improved, and good implementation and robustness are achieved.

The invention discloses a Chinese text category recognition system and method for unbalanced data sampling. The method comprises the following steps: firstly employing a text encoder to encode a Chinese text, and forming sentence vector representation; secondly, redundant and noise data are removed by adopting a multi-time DBSCAN clustering algorithm for negative sample data, and a deep learning sample selector is constructed for positive sample data to filter samples and select high-quality sample data to supplement positive samples while the samples are randomly generated; and finally, decoding the selected sample data through a text decoder to form text data, and forming a balanced data set together with the processed negative sample data so as to be applied to a text classification model. According to the method, a mixed sampling method is adopted to process data, particularly redundant data processing is performed on negative samples, a deep learning sample selector is constructedfor positive samples to filter generated samples, a high-quality sample training classifier is selected, and the text classification performance is improved.

The invention discloses a millimeter-wave radar environment map construction method and device based on a clustering algorithm. The method comprises the steps that: multi-frame mobile acquisition dataand corresponding pose information of a millimeter-wave radar are acquired; according to the pose information, the coordinate information of the target of the multi-frame mobile acquisition data relative to the radar in a Cartesian coordinate system is processed to obtain combined point cloud data, and the combined point cloud data comprises a plurality of sample vectors; a DBSCAN clustering algorithm is used for carrying out miscellaneous point screening processing on the combined point cloud data, and input parameters of the DBSCAN clustering algorithm are determined according to dimensioninformation of the combined point cloud data and Mahalanobis distance data between sample vectors in the combined point cloud data; a millimeter-wave radar measurement precision model is established by utilizing the combined point cloud data subjected to miscellaneous point screening processing; and a millimeter-wave radar environment map is constructed according to the millimeter-wave radar measurement precision model. The method can effectively improve the accuracy of the millimeter-wave radar environment map.

The invention discloses a meta-learning-based vehicle trajectory clustering method and system, belongs to the technical field of vehicle trajectory clustering, and solves the problem that an optimal clustering result cannot be obtained due to the fact that multiple different types of trajectory data adopt a single trajectory clustering algorithm in the prior art. A meta-learning-based vehicle trajectory clustering method comprises the following steps: collecting different types of GPS vehicle trajectory data, and obtaining an optimal DBSCAN clustering algorithm corresponding to the different types of GPS vehicle trajectory data; collecting GPS vehicle trajectory data, and obtaining a meta-learning device used for vehicle trajectory type division through training; and acquiring a vehicle trajectory type corresponding to the GPS vehicle trajectory data by using the meta-learning device, and clustering the GPS vehicle trajectory data by using an optimal DBSCAN clustering algorithm corresponding to the vehicle trajectory type to obtain a clustering result of the GPS vehicle trajectory data. The optimal clustering result can be obtained for various different types of trajectory data.

The embodiment of the invention discloses a geographic position locating method for dynamic IP multi-regional distribution. Clustering is performed by using a DBSCAN clustering algorithm based on the historical distribution condition of dynamic IP within a period of time to obtain a distribution range (the coverage is described by a circle) of a real position of the dynamic IP, and the distribution range of the IP is regarded as a locating result of the dynamic IP. Therefore, the inaccuracy caused by a single-point IP locating manner of an IP locating product in the prior art can be overcome; and the locating precision of the single-point IP locating manner can reach country, province, city, and even street, but the single-point IP locating manner cannot objectively reflect the real geographic position of the multi-regional dynamic IP.

A JSON data visualization optimization method in a cloud computing mode comprises the following steps: 1, constructing a target domain corpus; 2, carrying out corpus-oriented entity extraction; 3, incombination with Word2vec, performing instructive secondary pre-grouping on the corpus, and constructing a knowledge graph by using a DBSCAN clustering algorithm; 4, constructing a visual model tree VT; 5, defining a metadata tree MDT and an MST meta-structure tree, and converting JSON data returned by the REST service into MDT; and 6, querying whether data fragments matched in the knowledge graphconstructed in step 3 have actual semantic association or not on the basis of structure matching during pruning operation, dimension reduction adjustment and the matching process of StreuctModeli inan original structure tree MST and StreuctModeli in a visual model tree, and selecting an effective dimension combination according to an inquiry result optimization matching algorithm to improve theaccuracy of automatically generating the graph.

The invention discloses an image clustering method and system based on local chromatic features. A SURF-chromatic feature vector is extracted from each training image, K mean value clustering is carried out on the SURF-chromatic features of a whole training image set to obtain a feature vector dictionary; the SURF-chromatic feature vector is extracted from all images to be clustered, and a corresponding BOW feature vector is generated according to the feature vector dictionary; and DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering based on SNN (Shared Nearest Neighbor) similarity is carried out on the BOW feature vectors of all images to be clustered, and the images are subjected to corresponding category division according to the clustering. The method provided by the invention fully utilizes the local gray scale gradient and chromatic features of the image, has a better representational capability on the features of the image, and improves a clustering effect. In addition, when the images are clustered, the DBSCAN clustering method based on the SNN similarity can be used for processing clusters of different sizes, shapes and densities, and an image clustering effect can be improved.

The invention discloses a C-DBSCAN-K clustering algorithm under a Hadoop platform. The algorithm comprises the following steps of step 1.establishing clusters capable of communicating with each other; step 2.establishing the Hadoop platform for the clusters; step 3.using a dfs-put command to upload a to-be-clustered data set A to an HDFS; step 4.executing a Canopy clustering algorithm to carry out initial clustering on data in the A in order to obtain a clustering result of coarse granularity; step 5.constructing a k-d tree on the clusters obtained in the step 4; step 6.executing a DBSCAN algorithm to the clusters obtained in the step 4, using the k-d tree to query an epsilon-neighborhood of a data object in each cluster and outputting a clustering result; and step 7.merging the clusters with the same data in the step 6 and outputting a clustering result. The algorithm of the invention solves a problem of low clustering efficiency of the DBSCAN clustering algorithm on a large-scale data set in the existing technology.

The invention relates to the technical field of intelligent terminals, and especially relates to a path planning method and system based on scene classification. The method comprises the steps: obtaining a sample data set X; selecting a core sample data set M from the sample data set X; carrying out the cluster dividing of the core sample data set M, and obtaining a cluster dividing set C; inputting the cluster dividing set C into a corresponding navigation system, and carrying out the path planning; and outputting a path planning result. Each cluster in the cluster dividing set C is correspondingly taken as one scene class. According to the invention, the method is based on a DBSCAN clustering algorithm, and achieves the dividing of the sample data, i.e., the position data. The method achieves the optimization of a neighbor parameter calculating method in a conventional clustering algorithm, and an Epsilon and MinPts model is built through a core function, so as to achieve the correctand effective dividing of the clusters in a clustering process. When a central point of one cluster is selected and inputted into the navigation system, the method can effectively achieve the recommendation of the optimal path.

The invention discloses a trunk image-oriented open set recognition method, which comprises the following steps of (1) designing a CCN model, training the CNN model by adopting a part of training samples, and constructing a feature extractor; (2) for the feature maps extracted by the feature extractor, calculating the similarity between the feature maps, and clustering the feature maps by adoptinga DBSCAN algorithm; (3) designing a Loss function according to the similarity between the feature maps and the clustering result of the feature maps, and optimizing the parameters of the feature extractor and the weight parameters of the similarity function according to the Loss function; and (4) inputting the trunk images to be classified into an optimized feature extractor, calculating the similarity between the outputted feature images by using the optimized similarity function, and finally obtaining a DBSCAN clustering result of the feature images according to the calculated similarity value. According to the open set recognition method, the open set identification of the trunk images of unknown categories can be well realized.

The invention discloses an efficient monitoring method for the multi-ship collision risk in the sea area. The method comprises steps that instantaneous AIS data of all the ships in the monitored sea area are read, latitude and longitude coordinates of the ships are linearly interpolated, the position information of each ship is synchronized to the same time point; a scanning radius of DBSCAN is set according to monitoring requirements of the ship collision risk, and the minimum inclusion point number is set to be 2; DBSCAN clustering calculation is performed for all the ships in the sea area,and the ships in the sea area are clustered according to spatial distribution; and the multi-ship collision risk is calculated for each ship in the cluster generated through clustering. The method isadvantaged in that the total calculation of the collision risk of all the ships in the sea area can be reduced, the total calculation time is saved, and real-time calculation of the collision risk isrealized.

The invention discloses a real-time incremental and adaptive clustering method based on automobile radar data, in order to solve the problem of low clustering efficiency and inability to cope with theuneven data density clustering problem, a real-time clustering algorithm based on EKF and DBSCAN is proposed for automotive radar data in multi-target complex environment. The method of the inventiontakes into account the characteristics of EKF which is often used by automobile radar in tracking and predicting targets, improves DBSCAN algorithm, and the improved DBSCAN algorithm can ensure thatthe clustering result is not affected by track coincidence to a great extent. Moreover, the Kalman filter parameters can be continuously iterated in the same target, which saves the time needed from the initial parameter iteration and improves the clustering efficiency. The method of the invention simultaneously realizes the incremental and adaptive DBSCAN clustering, can keep lower time memory overhead, and can be used for solving the situation of uneven density of automobile radar data clusters.