853 results about "Feature description" patented technology

The present invention discloses a visual target detection and labeling method. The method includes: an image inputting step, to input an image to be detected; a candidate region extracting step, to extract a candidate window as the candidate region from the image to be detected using selectively search algorithm; a feature description extracting step, to perform feature description on the candidate region using a pre-trained large-scale convolutional neural network and output the feature description of the candidate region; a visual target predicting step, to predict the candidate region based on the feature description of the candidate region using a pre-trained object detection module, to estimate regions having the visual target; and a position labeling step, to labeling the position of the visual target according to the estimated result. Experiments show that, compared with the mainstream week supervision visual target detection and labeling method, the present invention has a stronger ability to excavate positive samples and a more general application prospect, and is suitable for visual target detection and automatic labeling tasks on the large-scale data set.

The invention relates to a point cloud automatic registration method based on normal vector. According to the method, processing objects are two or more than two pieces of three-dimensional point cloud data, wherein overlapped part exists every two pieces of adjacent three-dimensional point cloud data. The method comprises the following processing steps that (1) feature points are selected according to the point cloud local normal vector changes; (2) the histogram feature quantity is designed for carrying out feature description on each obtained feature point; (3) the initial matching dot pair is obtained through comparing the histogram feature vector of the feature points; (4) the precise matching dot pair is obtained through applying the rigid distance constraint condition and combining a RANSAC (random sample consensus) algorithm, and in addition, the initial registration parameters are obtained through calculation by using a four-element method; and (5) an improved ICP (iterative closest point)) algorithm is adopted for carrying out point cloud precise registration. The point cloud can be automatically registered according to the steps. The method has the advantages that feature description is simple, identification degree is high, higher robustness is realized, and registration precision and speed are improved to a certain degree.

The invention provides a robot semantic SLAM method based on object instance matching, a processor and a robot. The robot semantic SLAM method comprises the steps that acquring an image sequence shotin the operation process of a robot, and conducting feature point extraction, matching and tracking on each frame of image to estimate camera motion; extracting a key frame, performing instance segmentation on the key frame, and obtaining all object instances in each frame of key frame; carrying out feature point extraction on the key frame and calculating feature point descriptors, carrying outfeature extraction and coding on all object instances in the key frame to calculate feature description vectors of the instances, and obtaining instance three-dimensional point clouds at the same time; carrying out feature point matching and instance matching on the feature points and the object instances between the adjacent key frames; and performing local nonlinear optimization on the pose estimation result of the SLAM by fusing the feature point matching and the instance matching to obtain a key frame carrying object instance semantic annotation information, and mapping the key frame intothe instance three-dimensional point cloud to construct a three-dimensional semantic map.

The invention discloses a target automatically recognizing and tracking method based on affine invariant points and optical flow calculation, which comprises the following steps: firstly, carrying out image pretreatment on a target image and video frames and extracting affine invariant feature points; then, carrying out feature point matching, eliminating mismatching points; determining the target recognition success when the feature point matching pairs reach certain number and affine conversion matrixes can be generated; then, utilizing the affine invariant points collected in the former step for feature optical flow calculation to realize the real-time target tracking; and immediately returning to the first step for carrying out the target recognition again if the tracking of middle targets fails. The feature point operator used by the invention belongs to an image local feature description operator which is based on the metric space and maintains the unchanged image zooming and rotation or even affine conversion. In addition, the adopted optical flow calculation method has the advantages of small calculation amount and high accuracy, and can realize the real-time tracking. The invention is widely applied to the fields of video monitoring, image searching, computer aided driving systems, robots and the like.

A method for processing an anaglyph image to produce an enhanced image is described. The method includes receiving an anaglyph image; determining first and second feature locations from the first and second digital image channels and producing feature descriptions of the feature locations; and using the feature descriptions to find feature point correspondences between the first and second feature locations of the first and second digital image channels. The method further includes determining a warping function for the second digital image channel based on the feature point correspondences; producing an enhanced second digital image channel by applying the warping function to the second digital image channel; and producing an enhanced image from the first digital image channel and the enhanced second digital image channel.

The invention discloses a fast image splicing method based on an improved SURF algorithm. An existing corner extraction method and an existing corner feature description method are improved, mismatching of extracted corners is eliminated, and multiple images can be spliced fast. At first, an improved FAST algorithm is adopted for increasing the extracted corners, the operation speed of the FAST algorithm for extracting the corners is high, and after the improvement is carried out, the stability is good; secondly, the combination of SURF description and LBP description is adopted for describing the corner feature, and in this way, the speed for matching the corners can also be increased; then an RANSAC method is adopted for eliminating mismatching, accuracy is improved, and a more accurate transformation matrix is obtained, so that fast splicing is carried out; finally, according to obtained matching point pairs, parameters for transforming images to be spliced into reference images are calculated, and a slow-in and slow-out method is adopted for finishing image splicing.

In the development of an automatic speech recognition (ASR) system, an extensive study of the basic phonetic alphabet is performed to collect information regarding phonology and phonetics of the language or dialect in question (modern standard Arabic or MSA in this case). In addition, terminological and transcriptional problems are identified with respect to the language or dialect in question. Next, based on feature description (rather than symbol shapes), the symbols in the literature are mapped to a single or more recent phonetic alphabet. Lastly, from a maximal set containing all the phonemes, allophones, and transliteration symbols, a reduced set is created with a compact set of phonetic alphabets. Memory consumption is greatly reduced in a computer system by using this compact set of phonetic alphabets.

The invention discloses an optimized face recognition method and a device thereof, which improve the rate of facial image recognition. The technical proposal is that: the method and the device combine principal component analysis and linear discriminant analysis to solve the recognition problem; namely, the principal component analysis is carried out first before the conduction of the linear discriminant analysis so as to obtain relatively low dimensional space, then the space is utilized to carry out the linear discriminant analysis, and therefore, an intra-class dispersion matrix can not be caused and the process of the linear discrimination is effective. The invention firstly adopts the principal component analysis to obtain the best feature description, based on which, optimum identifying features are obtained by adopting the linear discriminant analysis, thereby greatly reducing the space dimension of facial features; finally, a minimum distance method is adopted to carry out classification and identification and therefore the rate of recognizing human faces is evidently increased. The method and the device of the invention are applied to face recognition.

The invention discloses an engineering drawing material information extraction method based on a template, comprising the following steps of: generating a table figure, words and filling rule description information of table units by using figure software to generate a table extraction template; reading and identifying basic figure element type information, figure property parameter information, rule description information and topological structure information which are contained in the extraction template; analyzing the feature of the extraction template to form table feature description according to the topological structure information; circularly reading and identifying basic figure element types and figure property parameter information in a CAD (Computer Aided Design) design drawing and then identifying table frames according to table features to form table frame integrations; circularly identifying the element of each table frame integration and then reading and identifying the basic figure element types and the figure property parameter information; and extracting material information and storing the material information in a database. The invention improves the extraction precision of the table features and ensures the extracted semantic relevance and the extraction accuracy of the material information.

The invention discloses a real-time robust far infrared vehicle-mounted pedestrian detection method. The method comprises the steps of catching a potential pedestrian pre-selection area in an input image through a pixel gradient vertical projection, searching an interest area in the pedestrian pre-selection area through a local threshold method and morphological post-processing techniques, extracting a multi-stage entropy weighing gradient direction histogram for feature description of the interest area, inputting the histogram to a support vector machine pedestrian classifier for online judgment of the interest area, achieving pedestrian detection through multi-frame verification and screening of judgment results of the pedestrian classifier, dividing training sample space according to sample height distribution, building a classification frame of a three-branch structure, and collecting difficult samples and a training pedestrian classifier in an iteration mode with combination of a bootstrap method and an advanced termination method. According to the real-time robust far infrared vehicle-mounted pedestrian detection method, not only is accuracy of pedestrian detection improved, but also a false alarm rate is reduced, input image processing speed and generalization capacity of the classifier are improved, and provided is an effective night vehicle-mounted pedestrian-assisted early warning method.

The invention discloses a method and a device for classifying images on the basis of a convolutional neural network. The method includes receiving various categories of inputted image samples and computing a neural network weight corresponding to each category of images; distributing the neural network weights corresponding to the various categories of images by the aid of a layered structure and forming a corresponding learning library in each layer; processing data of an inputted test category of image samples to obtain corresponding one-dimensional feature description, and performing feed-forward learning on the one-dimensional feature description corresponding to the data of the test category of image samples and the neural network weights in the learning libraries so as to judge whether the test category is available in the learned categories of images or not. The method and the device have the advantages that the problem of limitation of the traditional convolutional neural network on classification numbers can be solved by the layered distribution structure, the problem of excessive learning of the convolutional neural network can be solved, the classification capacity of the convolutional neural network can be expanded, the classification accuracy can be improved, and an image classification algorithm in new environments is high in robustness.

The invention discloses an image search method and apparatus. The image search method comprises the steps of obtaining a target region of interest of a to-be-searched image; extracting a local eigenvector and a deep learning eigenvector of the target region of interest; executing feature dimension reduction processing on the local eigenvector and the deep learning eigenvector by correspondingly utilizing a preset locally-weighted index and a preset depth-weighted index respectively, and performing feature fusion on the local eigenvector subjected to the dimension reduction and the deep learning eigenvector subjected to the dimension reduction by utilizing a preset splicing weighted index, thereby obtaining a target eigenvector which improves the feature description precision of the target region of interest; and performing a search according to the target eigenvector, thereby obtaining a search result based on the to-be-searched image.

The invention discloses an image retrieval database generating method, an image retrieval method, a corresponding image retrieval database generator, an image retrieval device and an image retrieval system. When a database is generated, sample image feature description information is classified and dimensionally reduced, and a label datum is formed for each feature point. Sample image content data and label data are stored in the database in a classified mode. During image retrieval, a target image to be retrieved is processed through a similar method, and a plurality of labels are formed for each feature point. During matching, traversal matching is carried out on the labels of the target image and all labels in a corresponding classified index in the database, and a matching value of the target image and sample images in the database is calculated. According to the image retrieval database generating method, the image retrieval method, the corresponding image retrieval database generator, the image retrieval device and the image retrieval system, the description content is compressed, and data transmission efficiency can be effectively improved; in addition, due to the fact that the sample image content data and the label data are stored in the classified mode, the image retrieval waiting time can be effectively shortened, and real-time retrieval can be possibly achieved in the large database.

An intelligent storage cabinet is formed by a box of a metal structure, an intelligent control computer, an identity and authority limit recognition system, a cabinet door unlocking control system and the like. Multiple feature description information requirements are given. Based on description information, operating steps of multiple operating modes are provided and include common user authorization, administrator authorization, common user authorization authority limit withdrawing/correcting, administrator authorization authority limit withdrawing/correcting, user storage operation, user fetching or renewing operation, user entrustment operation, user entrustment authority limit withdrawing/correcting, entrustment group setting, entrustment group withdrawing/correcting operation, user entrusted storage operation, user entrusted article fetching or renewing operation, storage cabinet cleaning or testing operation and user service receiving or refusing operation. A safety intelligent cabinet system for article/express delivery and a method thereof solve the following problems of the user authentication problem, the user use authority limit problem, the user authority limit relation problem, the authorization problem, the authority limit entrustment problem, the condition entrustment problem, and recycling problem, the casual user problem, the cabinet distribution strategy problem, the multi-cabinet storage problem, the management mode problem and the cabinet cleaning and detecting problem. Meanwhile, multiple improvement strategies are provided according to multiple special conditions, based on the set special description information and the set operating methods and according to certain special conditions.

The invention discloses a rapid image registration method based on sub-image corner features. The method includes the specific steps: firstly, selecting a reference sub-image and a to-be-registered sub-image, selecting one sub-image from a reference image as the reference sub-image, and selecting one sub-image with the same coordinate space as the reference sub-image from a to-be-registered image as the to-be-registered sub-image; secondly, extracting corners of the reference sub-image and the reference sub-image; thirdly, performing feature description on the corners extracted from the reference sub-image and the reference sub-image to obtain a feature vector of each corner; fourthly, performing similarity measurement and feature matching on the feature vectors of the corners on the reference sub-image and the reference sub-image to obtain K matching point pairs; and fifthly, adopting a least square method to compute a transformation matrix H between the reference image and the to-be-registered image based on the K matching point pairs, and registering the to-be-registered image onto the reference image based on the transformation matrix H. By the method, the requirement for image matching precision can be met, and image matching speed is increased greatly.

A descriptor generation unit 81 uses a first template prepared in advance to generate a feature descriptor, which generates a feature that may affect a prediction target from a first table including a variable of the prediction target and a second table. A feature generation unit 82 generates the feature by applying the feature descriptor to the first and second tables. A feature explanation generation unit 83 generates a feature explanation about the feature descriptor or the feature on the basis of a second template. An accepting unit 84 accepts values to be assigned to the first and second templates. The descriptor generation unit 81 generates the feature descriptor by assigning the accepted values to the first template, and the feature explanation generation unit 83 generates the feature explanation by assigning the values assigned to the first template to the second template.

The invention discloses an indoor human body behavior recognition method. The method comprises the following steps that: human body three-dimensional skeleton information is obtained based on Kinect equipment; three-dimensional skeleton features in each video set are extracted; the three-dimensional skeleton features are trained, and the features are described, and the training of the three-dimensional skeleton features further includes the following steps that: online dictionary learning is performed on the features, and then, sparse principal component analysis is performed on the features, and finally, a multi-task large margin nearest neighbor algorithm and a linear support vector machine are utilized to classify the features, so that a training feature set can be obtained; three-dimensional skeleton features of test videos are extracted; and the multi-task large margin nearest neighbor algorithm and the linear support vector machine are utilized to classify the features, so that feature descriptions can be obtained, and optimum judgment is performed on the training feature set and the test features with a scoring mechanism. The indoor human body behavior recognition method of the invention has a bright application prospect in intelligent video surveillance, patient monitoring systems, human-computer interaction, virtual reality, smart home, intelligent security and prevention and athlete assistant training, and has high feasibility and great social economic benefits.

The present invention provides an urban road network jam feature description analysis method based on data visualization. Generating a vehicle driving track sequence T aiming at the collected intelligent card port original detection data; performing visualization description analysis of the road network traffic operation condition based on the vehicle driving track sequence T, generating vehicle passing track-road section matching condition to perform statistics of the number of vehicle passing at the road section and overlap the display layers of the number of the vehicle passing on the electronic map, and performing road network jam state basic description analysis; and realizing the road network jam and traffic operation state analysis through the interaction operation. The urban road network jam feature description analysis method based on data visualization can allow the traffic managers to obtain clear traffic management control thinking through the road network jam time-space features and the described analysis of the road shunt features so as to provide basis for the research and the enforcement of the traffic signal control strategy and the path induction strategy.

The invention discloses an improved scale invariant feature transform (SIFT) image feature matching algorithm. The algorithm comprises: step one, scale space extreme points are detected; step two, feature descriptor is generated; and step three, a K-d tree balanced binary tree is built, a nearest neighborhood feature point on the K-b tree is searched by BBF, a matched feature dot pair is judged by Euclidean distance, and secondary matching is conducted after Euclidean distance matching. According to the improved SIFT image feature matching algorithm, the descriptor of 128 dimensions is reduced to 48 dimensions, execution efficiency of the algorithm is improved by two-thirds and reaches the speed of speeded-up robust features (SURF) feature description subalgorithm based on integral, and the defects that the algorithm is not suitable for the gray level and the changing circumstances of the point view of the images are overcome.

The invention relates to a preceding vehicle following method based on a deep convolutional neutral network and solves the problems that a self-adaptive cruise system in the prior art has tedious visual ranging and controller design processes, cannot fine adjust control parameters according to driving habits of drivers and cannot meet different personal driving habits. Driving behavior data are trained, and under the condition of given visual sensing input, accurate control of vehicle accelerators and automatic pedals under multiple work conditions is realized by imitating human drivers according to feature description of the surroundings by the deep convolutional neutral network. All that is required is to fine adjust the deep neutral network parameters with different work condition training samples instead of designing the controller structure and adjusting the controller gain under different work conditions. The tedious visual ranging and controller design processes are avoided, and the vehicle following behavior of the drivers is simulated by adjusting the deep neutral network parameters.