571results about How to "Guaranteed recognition accuracy" patented technology

The invention discloses an end-to-end identification method for a scene text with a random shape. The method comprises the steps of extracting a text characteristic through a characteristic pyramid network for generating a candidate text box by an area extracting network; adjusting the position of the candidate text box through quick area classification regression branch for obtaining more accurate position of a text bounding box; inputting the position information of the bounding box into a dividing branch, obtaining a predicated character sequence through a pixel voting algorithm; and finally processing the predicated character sequence through a weighted editing distance algorithm, finding out a most matched word of the predicated character sequence in a given dictionary, thereby obtaining a final text identification result. According to the method of the invention, the scene texts with the random shape can be simultaneously detected and identified, wherein the scene texts comprisehorizontal text, multidirectional text and curved text. Furthermore end-to-end training can be completely performed. Compared with prior art, the identification method according to the invention has advantages of obtaining advantageous effects in accuracy and versatility, and realizing high application value.

The present invention discloses a driving model training method, a driver identification method, a driving model apparatus, a driver identification apparatus, a device, and a medium. The driving modeltraining method includes the following steps that: the training behavior data of a user are acquired, wherein the training behavior data are associated with a user identifier; training driving data associated with the user identifier are obtained on the basis of the training behavior data; positive and negative samples are obtained from the training driving data on the basis of the user identifier, and the positive and negative samples are divided into a training set and a test set; the training set is trained by using a bagging algorithm, so that an original driving model can be obtained; and the test set is adopted to test the original driving model, so that a target driving model can be obtained. With the driving model training method adopted, the generalization of the driving model can be effectively enhanced; the problem of poor recognition results of current driving recognition models can be solved; and the accuracy of identifying the driving of drivers is improved.

The invention discloses a vehicle identification method and device and a vehicle. The method comprises the steps that a first image and a second image are acquired, the first image is a color image or a brightness image, and the second image is a depth image; road lane lines are acquired according to the first image; the road lane lines are mapped to the second image according to the interleaving mapping relation between the first image and the second image so that a vehicle identification range is generated in the second image; and the vehicle is identified according to the vehicle identification range. According to the method, the opposite-side vehicle can be rapidly and high-precisely identified, and alarm information can be rapidly and reliably generated on the basis of the identified opposite-side vehicle so that valuable braking reaction time can be won for the driver of the vehicle or the opposite-side vehicle and thus driving safety can be further enhanced.

The invention discloses an identity identification method based on self-established sample libraries and composite characters in video monitoring. The method comprises the steps that firstly, preprocessing is conducted on an acquired video, foreground detection is conducted, so that moving object information is obtained, then face detection is conducted on the basis of the moving object information, the detected face is identified, if the currently detected face can not be identified, a user is inquired so as to identify the detected face, and the identified face is added to the face sample library; if the face can not be detected in foreground information, pedestrian detection is conducted, a detected pedestrian is traced, gait period detection is conducted on a traced pedestrian image sequence, features of detected gait information of a period are extracted and identified, if identification fails, gaits are classified in the same mode of user identification and added to the gait sample library. The identity identification method provides a solution for identity identification on the condition of a lack of training sample diversity or small samples.

The invention discloses an expression identification method fusing a depth image and multi-channel features. The method comprises the steps of performing human face region identification on an input human face expression image and performing preprocessing operation; selecting the multi-channel features of the image, extracting a depth image entropy, a grayscale image entropy and a color image salient feature as human face expression texture information in the texture feature aspect, extracting texture features of the texture information by adopting a grayscale histogram method, and extracting facial expression feature points as geometric features from a color information image by utilizing an active appearance model in the geometric feature aspect; and fusing the texture features and the geometric features, selecting different kernel functions for different features to perform kernel function fusion, and transmitting a fusion result to a multi-class support vector machine classifier for performing expression classification. Compared with the prior art, the method has the advantages that the influence of factors such as different illumination, different head poses, complex backgrounds and the like in expression identification can be effectively overcome, the expression identification rate is increased, and the method has good real-time property and robustness.

The invention discloses a scene text end-to-end recognition method based on boundary point detection, and the method comprises the steps: extracting text features through a feature pyramid network, and generating candidate textboxes through a region extraction network; detecting a more accurate multi-directional bounding box of the text instance through a multi-directional rectangular detection network; secondly, detecting an upper boundary point sequence and a lower boundary point sequence of the text in the multi-directional bounding box; and finally, converting the text in any shape into ahorizontal text by utilizing the detected boundary point sequence for the subsequent attention mechanism-based sequence recognition network to performing recognizing, and finally, finding out the mostmatched word of the prediction sequence in the given dictionary by utilizing a cluster search algorithm to obtain a final text recognition result. According to the method, the scene text in any shapein the natural image can be detected and recognized at the same time under the condition that character-level labeling is not needed, the scene text comprises the horizontal text, the multi-directiontext and the curved text, and end-to-end training can be completely carried out.

The invention discloses an identity authentication method. The method comprises the following steps: acquiring a face image of a user to be identified through a composite imaging device; performing human eye detection on the face image, and if human eyes are detected in a second area image, entering a face identification mode; determining whether the second area image satisfies a face identification condition; if so, performing face identification; according to a face identification result, preliminarily determining whether the user is legal; if the user is preliminarily determined as a legal user, entering an iris identification mode; re-obtaining a first area image, including the human eyes, of the user in a first area; positioning and cutting an eye image from the first area image, and determining whether the eye image satisfies an iris identification condition; if so, performing iris identification; and according to an iris identification result, re-determining whether the user is a legal user. The invention further discloses a corresponding identity authentication device and a mobile terminal.

The invention discloses a self-adaptive automobile instrument detection method based on character segmentation cascaded quadratic classifier, relating to the automobile instrument pointer visual detection technology field. In order to solve problems of a current instrument detection system in the automobile instrument detection field, the self-adaptive automobile instrument detection method basedon character segmentation cascading performs threshold segmentation on an original image and performs analysis processing on morphology and a connected domain, adopts a method based on couture analysis method to finely extract pointers, establishes a pointer information list, constructs a character segmentation cascading quadratic classifier consisting of an HOG / SVM quadratic classifier, a character filter and a CNN digital classifier which are connected in a cascaded mode, adopts the cascaded quadratic classifier to identify a number character area of the instrument, performs local analysis to extract scale points with the number character area as a center, determines a position of an angle corresponding to the scale point, establishes a Newton interpolation linear description relation between the instrument pointer angle and a response value and determines whether the instrument is qualified. The self-adaptive automobile instrument detection method based on the character segmentationcascaded quadratic classifier is applied to the automobile instrument pointer vision detection field.

The invention provides an individual recognition method and device and a program. The method comprises the steps that candidates, with matching degrees higher than a first critical threshold in regard to a first biological feature, in a database are determined as a candidate set; the candidate, with the matching degree higher than a first threshold in regard to the first biological feature, in the candidate set is determined as an individual, and recognition is exited; when all the matching degrees, in regard to the first biological feature, of the candidates in the candidate set are not higher than the first threshold, the candidate cannot be judged in the first judgment part, the candidate, with the matching degree higher than a second critical threshold in regard to a second biological feature, in the candidate set is determined as the individual, and recognition is exited; and when the candidate set is empty, the candidate, with the matching degree higher than a second threshold in regard to the second biological feature, in the database is determined as the individual, and recognition is exited, wherein the first threshold is higher than the first critical threshold, and the second threshold is higher than the second critical threshold. Through the method, a high passing rate and high precision of face recognition and anti-fake capability of palm recognition are achieved at the same time, matching time is shortened while accuracy is ensured, and recognition efficiency is improved.

The present invention discloses a speech recognition method and system. The method comprises: obtaining the special type of information text from a pre-determined data source; performing statement segmentation of each obtained information text to obtain a plurality of statements, performing segmentation processing of each statement to obtain corresponding participles, forming a first mapping corpuses through adoption of each statement and the corresponding participles; and according to each obtained first mapping corpus, training the preset type of a first language model, and performing speech recognition based on the trained first language model. The precision of speech recognition is effectively improved, and the cost of speech recognition is effectively reduced.

The invention discloses a target detection method, a target detection model and a target detection system based on a cascade detector. The target detection method comprises the following steps of S1,training the target detection model by adopting a training data set with a target label; S2, inputting a to-be-detected picture into the trained target detection model, and performing feature extraction, upsampling and feature fusion to obtain feature maps P2, P3,..., Pn with different sizes; S3, predicting according to the feature map Pn to obtain a target position Bn and a category Cn; extracting corresponding features from the corresponding feature map Pi-1 according to the target position Bi for target prediction to obtain a target position Bi-1 and a category Ci-1, i = 3-n; S4, taking thetarget position B2 as a final prediction result. According to the invention, through the multiple iterative regression prediction of the the target candidate box, the target position with more accurate prediction is obtained, the target detection precision is improved, and the detection accuracy is higher.

The invention relates to a pedestrian re-identification method fusing random batch masks and multi-scale representation learning. The pedestrian re-identification method comprises the steps of constructing a pedestrian re-identification training network; performing network hyper-parameter adjustment according to preset training parameters to obtain a learning network; shielding multi-scale representation learning and random batch mask branches to obtain a test network, and inputting the test set into the test network to obtain a corresponding test identification result; judging whether the accuracy of the test recognition result is greater than or equal to a preset value or not, if so, inputting the actual data set into the learning network, and otherwise, retraining the network; and finally, shielding multi-scale representation learning and random batch mask branches to obtain an application network, and inputting the query image into the application network to obtain a correspondingidentification result. Compared with the prior art, the method has the advantages that a random batch mask strategy, multi-scale representation learning and loss function joint training are used, moredetailed discrimination features of pedestrian images can be captured, and local important suppressed features are extracted.

The invention discloses a large-scale face identification method based on GPU accelerated retrieval, which relates to the field of computer vision. The large-scale face identification method comprisesthe steps of face detection and alignment, face feature extraction, Hash feature acquisition, face index database establishment, multi-GPU accelerated Rough matching, hash-based candidate set acquisition, precise matching based on distance metric, voting to obtain a best-matched person and the like. The large-scale face identification method based on GPU accelerated retrieval is based on two-stage feature matching of Hash index and multi-GPU accelerated computing, can accelerate the screening of eigenvectors by utilizing the powerful parallel computing capacity of the GPU, greatly reduces time consumption on large-scale data set retrieval, and can meet various application requirements taking the implementation of a deep convolutional neural network as the basis and having high demand on real-time performance.

The invention discloses a transfer learning-based multi-view commodity image retrieval and identification method. The method comprises the steps of 1, establishing a multi-view image basic library according to a commodity list, performing fine adjustment on a pre-trained deep residual error network by using a small amount of commodity images through a transfer learning technology, extracting features of the image basic library by using the network, performing dimension reduction on the features, constructing a feature library, and finally according to corresponding relationships among the feature library, the image basic library and commodity types, establishing a mapping table; 2, after to-be-identified commodity images are obtained, extracting features of the images by using the networkand performing dimension reduction; and 3, performing distance measurement on the features of the to-be-identified commodity images and the features of the images in the basic library, taking the mostsimilar image with the shortest distance as a matching result, and through the mapping table, obtaining commodity type names of the to-be-identified commodity images. The features with strong representation capabilities can be automatically extracted; a semantic gap is further broken through; and the retrieval efficiency and the identification precision are improved by only utilizing a small amount of image basic libraries and low-dimensional features.

The invention discloses a model parameter verification method, device and apparatus in transverse federated learning and a medium, which are used for solving the problems of lower model training efficiency and accuracy caused by incapability of identifying invalid model parameters in the prior art. The method specifically comprises the steps that model parameters reported by all model training apparatuses are issued to other model training apparatuses except the model training apparatuses of the model training apparatuses through a server for error evaluation; the server can obtain an error value set of each model parameter; therefore, according to the error value set of each model parameter; screening invalid model parameters, therefore, in the model training process, when the initial model parameters used in the next model training period are determined, the invalid model parameters can be eliminated, so that the problem that the model training efficiency and accuracy are low due tothe fact that the invalid model parameters cannot be recognized is effectively solved, and the model training efficiency and accuracy are improved.

The invention relates to a laser scanning based vehicle type recognition method for free flow charging. The laser scanning based vehicle type recognition method includes the following steps: by a laser scanning detector, parallelly scanning two sections of a travelling vehicle to acquire measured angle and distance data, wherein the two sections are perpendicular to the road plane and intersect; by the laser scanning detector, uploading the measured angle and distance data to a vehicle type recognition device; by the vehicle type recognition device, calculating to acquire length, width and height of the vehicle and outline information of the whole vehicle; matching the acquired vehicle outline information with a vehicle type classification feature library to acquire a vehicle type to achieve automatic vehicle type recognition. By the laser scanning based vehicle type recognition method, automatic vehicle type recognition of all passing vehicles of free vehicle flows with the average vehicle speed below 120 kilometers / hour can be realized.

The invention relates to a human body behavior recognition method based on global characteristics and sparse representation classification. The method comprises the following steps: performing Gaussian kernel convolutional filtering preprocessing on a video frame, and extracting a moving foreground pixel by using a differential method; sampling a pixel value according to a time space dimension ofa parameter, determining a moving area, adjusting the size of the video frame, performing primary dimension reduction, splicing video frames in rows to form a vector group, and acquiring characteristic vectors; splicing the characteristic vectors in rows to form a characteristic matrix, performing secondary dimension reduction, calculating a primary characteristic dictionary of the characteristicmatrix, initializing the dictionary, after dictionary initialization, performing dictionary learning by using a class accordant K-time matrix singular value decomposition method, calculating an inputsignal sparse code according to the dictionary, inputting the code into a classifier, and outputting a behavior type; and counting dictionary learning parameters, and performing behavior recognition in real time. By adopting the method, dictionaries and linear classifiers with both reconstitution functions and classification functions are acquired, human body behavior recognition efficiency is improved, and the method is applicable to scientific fields such as security monitoring, video search based on contents and virtual reality.

The invention discloses a method for recognizing road rage states of drivers on the basis of electroencephalography and pulse information. The method includes steps of 1), acquiring pulse informationand electroencephalography signals of the drivers and preprocessing the pulse information and the electroencephalography signals; 2), extracting features of the pulse information and the electroencephalography signals of the drivers; 3), fusing electroencephalography features and pulse features of the drivers and reducing dimensions of the electroencephalography features and the pulse features; 4), training driver road rage state discrimination classifiers; 5), judging the road rage states of the drivers in real time, to be more specific, judging the road rage states of the drivers by the trained road rage state discrimination classifiers in real time and prompting the drivers when judgment results are 'yes'. The step 1) particularly includes (1), acquiring the pulse information of the drivers by wrist strap type terminals and preprocessing the pulse information; (2), acquiring the electroencephalography signals of the drivers by head-mounted terminals and preprocessing the electroencephalography signals. The step 2) particularly includes (1), extracting the features of the pulse information of the drivers; (2), extracting the features of the electroencephalography signals of the drivers. The method has the advantage that the problems of high power consumption and susceptibility to external conditions in the prior art can be solved by the aid of the method.

The invention provides an iris recognition method and apparatus, and a mobile terminal. The method comprises steps of determining a corresponding current intensity between a human eye and a camera to be a first current intensity according to the distance between the human eye whose iris is to be acquired and the camera; supplying power to an infrared light according to the first current intensity so that the infrared light emits infrared light; collecting an iris image of the human eye under irradiation of the infrared light; and carrying out iris recognition according to the iris image. The method can improve the accuracy of iris recognition.

The invention discloses a vehicle logo recognition method and system based on a selective search algorithm. The vehicle logo recognition method based on a selective search algorithm includes the steps: performing positioning of a license plate on an original vehicle image, and acquiring the position of the license plate; according to the position of the license plate, the spatial position relationship between the license plate and an vehicle logo, and the vehicle window edge information, performing coarse positioning on the vehicle logo in the original vehicle image, and obtaining an vehicle logo coarse positioning image; based on a central axis of the vehicle, selecting a vehicle logo candidate area in the vehicle logo in the original vehicle image; using a selective search algorithm to perform target positioning on the vehicle logo candidate area, and obtaining a positioning target set; utilizing a linear constraint coding algorithm to train a vehicle logo determination classifier to perform determination on the vehicle logo for the positioning target set to obtain the position of the vehicle logo; and utilizing the linear constraint coding algorithm to train a multi-type vehicle logo recognition classifier to perform concrete type recognition on the vehicle logo, and obtaining the vehicle logo recognition result. The vehicle logo recognition method and system based on a selective search algorithm have the advantages of extensive applicability, high robustness and high detection speed, and can be widely applied to the image processing field.

The invention discloses a user identity recognition method based on moving trajectory similarity comparison. Firstly, a city is divided into a longitude and latitude network, the frequency of trajectories appearing in each grid is counted, and grids with the number of points exceeding a threshold value Vp is mapped to a set; the similarity of two paired trajectories is calculated; finally a candidate threshold value Vr is set, namely Top-N sequences with the number of appearing areas exceeding Vr form a candidate set, and a trajectory Top-N sequence which is most similar to the trajectory of ato-be-identified object is matched in a pre-stored trajectory Top-N sequence set. Compared with a traditional method of directly calculating frequency distribution vectors, the areas which frequentlyappear have stronger regularity, so that the calculated amount is significantly reduced without sacrificing the accuracy of recognition.

The invention discloses a full-automatic three-dimensional conversion method aiming at a grating architectural plan. The method comprises the following steps of (1) performing binaryzation and correction on the grating architectural plan to obtain a preprocessed image; (2) extracting an image region comprising wall lines from the preprocessed image to obtain a plurality of sub-images; (3) performing vectorization processing on all sub-images to correspondingly obtain a line segment set with wall width, and performing extraction operation on the line segment set to obtain information such as wall positions and the wall width; (4) obtaining subgraphs of wall appendants through the line segment set in step (3), and judging the specific classes of the subgraphs of the wall appendants by using a linear identification analytical algorithm; and (5) converting data in steps (3)and (4) into three-dimensional structure data according to a preset height.

The invention discloses a fixed-lens real-time monitoring video feature extraction method based on SIFT feature clustering and the method comprises the steps: carrying out the feature extraction of each frame of a monitoring video, generated in real time, in a mode of parallel computing through employing an SIFT feature extraction algorithm; enabling the monitoring video stream generated in real time to be segmented into video segments according to the rule that each video segment comprises the same content; and respectively extracting a special key frame of the each video segment after segmenting. The method effectively separates the video segments with the similar contents from the monitoring video, effectively extracts the key frame from the similar video segments through employing a key frame extraction method based on a maximum feature point strategy, reduces the redundancy of the key frame, achieves the better video feature extraction effect, and provides a basis for the content retrieval of a large number of monitoring videos. Meanwhile, the method effectively solves a difficulty that the time cost of the feature extraction of video frames is large through enabling the processes of feature extraction of the video frames to be parallel, and improves the instantaneity.

The invention provides an intersection point feature extraction based digital identification method. The method comprises the following steps: (1) image preprocessing; (2) character zone positioning; (3) character segmentation; (4) intersection point feature extraction; and (5) character identification. According to the method, a noise zone is prevented from being misjudged to be a digit, so that the digital identification accuracy is improved; and under the condition of ensuring the identification precision and the anti-jamming property, the calculation amount is greatly reduced, the method has a relatively high application value for identifying a digital instrument in a power grid, and an instrument digital real-time identification system with a relatively good effect is realized. The method has the advantage that normalization and refinement processing does not need to be carried out in the processing method.

The invention discloses a grounding fault diagnosing method and apparatus of a DC 600V train power supply system. The method includes the steps of obtaining fault waveform sets of the DC600V train power supply system when different types of ground faults occur, wherein each fault waveform sample set corresponds to one ground fault type, training each fault waveform set to obtain a fault classification model, and acquiring current fault waveform on a real-time basis when a grounding fault occurs to a target DC600V train power supply system and diagnosing the grounding fault type corresponding to the current fault waveform according to the fault classification model. The apparatus comprises a classification model training module and a fault diagnosing module. The method and apparatus have the advantages of simple realization operation, high diagnosis efficiency and precision, high universality, good expansion performance and easy execution and maintenance.

The invention provides a construction method of a fabric defect recognition system based on a lightweight convolutional neural network. The method comprises the following steps: firstly, configuring an operation environment of a fabric defect identification system; obtaining a lightweight convolutional neural network according to factorization convolution; then, collecting fabric image sample data; standardizing the fabric image sample data; dividing the standardized fabric image sample data into a training image set and a test image set; inputting the training image set into a lightweight convolutional neural network for training by using an asynchronous gradient descent training strategy to obtain an LZFNet-Fast model, and finally inputting the test image set into the LZFNet-Fast modelfor testing to verify the performance of the LZFNet-Fast model. According to the method, a standard convolution layer is replaced by a factorized convolution structure, the colored fabric with complextextures is effectively identified, the number of parameters and the calculated amount of the model are reduced, and the identification efficiency is greatly improved.

The invention discloses an airborne visual detecting and multi-target positioning system of an unmanned gyroplane, and belongs to the technical field of positioning navigation and control. The system comprises an airborne subsystem and a ground monitoring subsystem; the airborne subsystem comprises a video collecting unit, an image processing unit and an image transmission transmitting terminal, and an image collected by the video collecting unit is processed by the image processing unit and then transmitted to the ground monitoring subsystem through the image transmission transmitting terminal; the ground monitoring subsystem comprises a ground station and an image transmission receiving terminal connected with the ground station, and the image transmission receiving terminal is communicated with the image transmission transmitting terminal. The system is compact in structure and achieves perfect integration with the unmanned gyroplane, flight and control of the unmanned gyroplane are convenient, therefore, the precision of target positioning is effectively guaranteed, and multi-target positioning is achieved. The invention further discloses an implementation method of the airborne visual detecting and multi-target positioning system of the unmanned gyroplane.

In some examples, processing circuitry obtains a shallow hash neural network (SHNN) model that has been trained from a HNN model and based on a set of SHNN training images that aggregates image recognition results from at least two reference hash neural network (HNN) models. Further, the processing circuitry performs an image recognition on the image according to the SHNN model, to obtain an image class vector in an image class space. The image class vector includes probability values of respective image classes in the image class space. A probability value of an image class in the image class space is a combination of intermediate probability values of the image class that are resulted from the at least two reference HNN models. Further, the processing circuitry determines one of the image classes for the image according to the probability values of the respective image classes in the image class space.

The invention provides a somatic action identifying method. The somatic action identifying method includes steps of 1) acquiring effective data segments of the somatic action by a speed sensor in the somatic device; 2) calculating DTW distance between the acquired effective data segments and the standard data segments and matching a sample library according to the DTW distance between the acquired effective data segments and the standard data segments; 3) calculating the DTW distance between the acquired effective data segments and each sample data segment in the matched sample library of the step 2), and using identifiers of the somatic action indicated by the minimum sample data segment of the DTW distance as the identifying results of the acquired effective data segments. The invention further provides a corresponding man-machine interaction device. On the premise of guaranteeing identifying accuracy, identifying efficiency is improved; high identifying accuracy is guaranteed on the premise that holding ways of the device during action identification are not limited.

The invention discloses a transient power disturbance identification method based on S conversion and an improved SVM algorithm. The method comprises the following steps: (1), carrying out processing on a disturbance signal based on improved S conversion; (2), extracting a disturbance signal characteristic; and (3) designing an SVM classifier based on a semi-supervised learning algorithm to classify samples. Compared with the previous power quality disturbance classification method, the provided method has beneficial effects: on the premise that the identification accuracy of the SVM algorithm is guaranteed, the improved semi-supervised learning algorithm is introduced into the sample with low reliability in the SVM algorithm, so that the identification accuracy of the disturbance signal can be improved; and advantages of good scientific and reasonable performances, high adaptability, and great promotional value and the like are realized.