68results about How to "Simple framework" patented technology

The present invention is directed toward a system and process that controls a group of networked electronic components using a multimodal integration scheme in which inputs from a speech recognition subsystem, gesture recognition subsystem employing a wireless pointing device and pointing analysis subsystem also employing the pointing device, are combined to determine what component a user wants to control and what control action is desired. In this multimodal integration scheme, the desired action concerning an electronic component is decomposed into a command and a referent pair. The referent can be identified using the pointing device to identify the component by pointing at the component or an object associated with it, by using speech recognition, or both. The command may be specified by pressing a button on the pointing device, by a gesture performed with the pointing device, by a speech recognition event, or by any combination of these inputs.

A visual target tracking method of a full convolution class and regression twin network structure comprises the following steps: (1) according to the position of a target in an image, cutting a targettemplate image and a search area image in an original training set, and forming a training data set by cut image pairs; (2) establishing a full convolution twin network to extract image features; (3)establishing a classification regression network; (4) in response to the fact that each pixel point on the image has a corresponding foreground score and a predicted bounding box, calculating the total score of each pixel point by combining the information of the foreground score and the information of the bounding box, wherein the pixel point with the highest total score is the center of the tracking target; and (5) training the full convolution twin network and the classification regression network by using the training data set to obtain the trained full convolution twin network and the classification regression network, calculating a score graph of a target in the to-be-tested image sequence by using the trained networks, and performing target positioning based on the score graph. According to the invention, the tracking precision and speed are improved.

Thorough optimization and improvement have been made on the basis of original closed-shelf stack room by the unattended full-automatic book management system of the invention. The concept of a book basket is brought in to process an order of each borrower in a batch. The device is simplified and the efficiency of single functional part is enhanced during the period of storing books, taking books, returning books, conveying a book basket and transmission, thereby the functions of the whole system being optimized. The system includes bookracks, a laneway vehicle, a laneway vehicle interface, a transmission device, a book taking window, and a book returning window, and further includes a book basket and a book box independent of the bookracks. The data information inside the database of the closed-shelf stack room system is administered and operated by the closed-shelf stack room management system to accomplish the interaction between the collection system and the control system of the full-automatic closed-shelf stack room. Ultimately books are administered through the full-automatic closed-shelf stack room.

The invention discloses a non-rigid multiscale object detection method based on a convolutional neural network, which comprises the steps of firstly designing a deep network, performing deep network learning and training for target detection by infrared pedestrian data sets with certain representativeness cvc-09, cvc-14 and self marked small target pedestrian data, wherein the infrared pedestriandata sets have small targets with the pixel being about 20*30 and big target pedestrians at a close range; then designing different detection models based on a hardware platform, and evaluating the detection models. The designed network applies 7*7 grid partition and applies six candidate boxes in allusion to each grid, target area nomination is accomplished, and pedestrian target detection is achieved through classification recognition and location regression analysis. The neural network is applied in the design, and the feature extraction ability of the detection model is increased. The real-time performance of detection is realized based on the idea of grid partition, the number of the candidate boxes is increased by applying data sets with different target scales and multiscale training, and finally the small target detecting ability is enhanced.

The invention discloses a visual design system of the artificial intelligence of a game on the basis of an XML (Extensive Markup Language). The system comprises a system logic device, a graph editor, a function editor, a graph configurator and an integral actuator, wherein the system logic device comprises a connection assembly and a function assembly and is used for describing assembly logics; the graph editor is used for graphically expressing the assemblies and combining the assemblies into an artificial intelligence tree structure; the function editor is used for compiling corresponding attributes and functional interface functions for the assemblies; the graphic configurator is used for configuring the attributes and the functional interface functions to the assemblies; and the integral actuator is used for traversing, backtracking, interrupting and blocking the assemblies of the artificial intelligence tree to cause the artificial intelligence tree structure to own an integrated function. The visual design system has the beneficial effects that a development period of the artificial intelligence is obviously shortened, the maintainability and the reusability of codes are enhanced, game experience is improved, personnel on various posts can work in cooperation with a due division of labor, and a flow is optimized.

A dual action weightlifting machine has a stationary main frame assembly comprising a right side frame and a left side frame on opposite sides of an exercise area, and a traveling frame movably supported on the stationary frame assembly and having right and left sides and a cross bar connecting the right and left sides above the exercise area, each side of the traveling frame having a vertical guide. A weight bearing exercise bar is movably supported on the traveling frame and has spaced first and second vertical slides slidably mounted on the right and left vertical guides. A horizontal slide assembly has dual horizontal guides on each side frame which engage dual slides on each side of the traveling frame. The horizontal slide assembly is located in a lower portion of the stationary frame assembly completely below the vertical travel path of a user engaging portion of the weight bearing exercise bar.

Modular human powered recumbent vehicle has a variable configuration and size. It comprises a group of standardized modules—fast attachable parts connected together by set of cross tubes (4) of variable length. It has three basic modules: middle (1), left (2) and right (3); each module has several modifications. One or more additional modules, like child seat (17), shopping basket (18), additional pedal (15) or electric (16) drive, also can be attached for performing a specific task. Vehicle can be transformed to a powered or unpowered trailer of variable configuration. It can be folded for storage and disassembled for transportation.

The invention discloses a video target tracking method based on pyramid convolution, and the method comprises the following steps: cutting out target template images and search region images corresponding to all images in an image sequence training set according to the position and size of a tracking target in the images; constructing a full convolution twin network based on pyramid convolution; training the full convolution twin network based on pyramid convolution by using the constructed image sequence training set to obtain a trained full convolution twin network; and carrying out single target tracking by using the trained pyramid convolution-based full convolution twin network. In the target tracking process, pyramid convolution is added on the basis of a traditional full convolutiontwin network, convolution kernels of different scales and depths are used for extracting and fusing multi-scale features, detail information of different levels in a scene is extracted, and then robustness and accuracy of a tracking system are improved.

The invention belongs to the technical field of artificial intelligence, and particularly relates to a domain adaptation method based on a machine vision scene text recognition task. The method comprises the following steps: constructing a CNN-LSTM network and an attention network; combining the CNN-LSTM network and the attention network into a scene text recognition network; inputting the scene images from the source domain and the target domain into a scene text recognition network, extracting image features from the input scene images by CNN-LSTM, recoding the image features by an attentionnetwork, extracting corresponding features of each character, and segmenting text information in the images into character level information; and finally, constructing a domain classification networkby applying a transfer learning technology based on adversarial learning, forming an adversarial generation network together with the scene text recognition network, and finally enabling the model toeffectively adapt to a target domain. According to the method, a small number of target domain calibration samples are fully utilized, the problem of sample scarcity frequently occurring in an actualscene text recognition task is solved, and the recognition effect is improved.

The invention relates to an image object co-segmentation method guided by local shape migration, comprising the following steps: inputting M images containing objects of the same semantic category, making a saliency analysis of each image, and generating a foreground-background initial segmentation result; carrying out dense feature point matching for any two images; establishing the corresponding relationship between each local image area and local areas of other images according to the matching result; learning the weight of the corresponding relationship using a local linear structure maintenance algorithm; and using an iterative algorithm to transfer the foreground-background segmentation result between the corresponding local areas to get a final segmentation result. The method has good performance in co-segmentation of image objects of the same semantic category, and can be applied to fields such as image content understanding and image object recognition.

The invention discloses a product supply and demand network interaction platform and method. The platform comprises a demand end, a background processing end, a supply end and a logistic end; through the demand end, a demand side can put forward demanded products, quantity and reception information of the products; the background processing end comprises an encryption processing module and an order forming module; the encryption processing module carries out encryption processing on the reception information so as to form reception two-dimensional codes; the order forming module forming the products and the quantity put forward by the demand side at the demand end into delivery lists and sending the delivery lists and the reception two-dimensional codes to a corresponding supplier; through the supply end, the supplier obtains a corresponding list and a corresponding reception two-dimensional code; after a logistic staff obtains the delivered products from the supplier, the logistic staff obtains the reception information by scanning the reception two-dimensional codes through the logistic end and then distribute the delivered products. The product supply and demand network interaction platform and method are simple in frame and convenient to operate, and can effectively ensure the information security of the demand side so as to improve the privacy degree of the demand side.

A magnetic field sensing apparatus including a substrate, first, second, and third magnetic field sensing units, and a switching circuit is provided. The substrate has a surface, and has a first inclined surface and a second inclined surface. The first magnetic field sensing unit includes a plurality of magnetoresistance sensors connected together to form a Wheatstone full bridge and disposed on the surface. The second magnetic field sensing unit includes a plurality of magnetoresistance sensors connected together to form a Wheatstone half bridge and disposed on the first inclined surface. The third magnetic field sensing unit includes a plurality of magnetoresistance sensors connected together to form a Wheatstone half bridge and disposed on the second inclined surface. The switching circuit electrically connects the second magnetic field sensing unit and the third magnetic field sensing unit. A magnetic field sensing module is also provided.

The invention discloses a zero-sample target detection system and a learnable semantic and fixed semantic fusion method, a zero-sample learning mechanism is introduced into a deep target detection framework, a set of zero-sample target detection system LATNet with strong discrimination capability is established, and an end-to-end zero-sample target detection task is realized through the LATNet. Alearnable semantic feature and fixed semantic feature combined method is used, so that when a network is trained in a source domain, word vector information of a category can be fully utilized, end-to-end learning can also be utilized, a category prototype with better identification capability is discovered, and the best detection accuracy is obtained. The system is simple in framework, convenientto use, high in expandability and high in interpretability, and the results of the two tasks of zero sample detection and generalized zero sample detection of the two mainstream visual attribute datasets exceed those of an existing method. And the support of a basic framework and a method is provided for the target detection technology in the military and industrial application fields.

The invention relates to a safety communication method of a wireless mesh network which belongs to the technical field of network. The method comprises the following steps: requesting initial authentication, starting an initial authentication protocol procedure by an authentication server and performing enciphered data communication by a pre-shared key and other legal wireless mesh network nodes in the wireless mesh network to realize functions of the wireless mesh network. The method meets new requirements of dynamic self-organization of the wireless mesh network, also provides safety performance approaching to 802.11 standard requirements for the network. The method is based on a wireless network communication standard of 802.11 protocol family, and is applicable to the wireless mesh network established based on 802.11 link. The method provides simple frame and structure, ensures compatibility and flexibility, and is easily realized.

A simple structure is provided for the frame by changing the orientation of the crank mechanism so as to orient the crankshaft longitudinally. More specifically, the structure includes two ring-shaped main plates are provided in the front and back and are connected by means of connector plates. An upwardly and downwardly movable slide for the machine press is provided in a space demarcated by four column portions of the ring-shaped main plates and a bolster is provided atop a bed of the ring-shaped main plates. A crankshaft oriented longitudinally is provided at the crown portion of the ring-shaped main plates, and the slide is moved upward and downward by a drive mechanism that includes the crankshaft.

The invention relates to the field of three-dimensional target detection and tracking, in particular to a three-dimensional object detection and tracking method based on streaming data, which comprises the following steps: inputting a front key frame and a rear key frame comprising point cloud data and image data, performing feature extraction on the data to obtain a feature map, and performing related operation on the two feature maps to obtain a related feature map; inputting the feature map into a candidate box extraction network to obtain a candidate box; and obtaining feature blocks fromthe feature map and the related feature map through the candidate boxes and inputting the feature blocks into the regression network to obtain a three-dimensional box and a three-dimensional box offset of the detected object respectively; solving other frames except the key frames through an interpolation method, and associating targets in all frames to obtain a tracking result. Only the key frames need to be detected, the stream data detection speed is increased, the requirement of an automatic driving environment for real-time performance is met, and meanwhile better stability is achieved; meanwhile, point cloud information and image information are fused, advantages and disadvantages are complementary, and the accuracy of object detection is improved.

A stereovision system is disclosed, which comprises: at least one diffractive optical element and an optical imaging device. Each of the diffractive optical element is used for allowing a first beam containing information relating to an object to pass through and thus transforming the same into a second beam containing information relating to the object. The optical imaging device is used for receiving the second beam so as to concentrate the energy thereof for forming an Mth-order diffraction image. By combining the aforesaid Mth-order diffraction image with another energy-concentrated Nth-order diffraction image, a series of images can be formed. Accordingly, by comparing the disparity between corresponding points in the series of images, the distance between the object and the diffractive optical element can be obtained. It is noted that the aforesaid M and N represent the order of diffraction.

The invention discloses a CA-NARX water quality prediction method based on meteorological factors, and belongs to the technical field of intelligent water quality prediction data application. The method comprises the following steps: 1, ; data standardization, 2, creating a sample matrix, 3, determining an initial clustering center according to the quantile; (4) initial clustering is carried out according to the Euclidean distance; (5) the mean value of each class is used as a new clustering center; (6) clustering is carried out in batches according to the Mahalanobis distance between each sample and the clustering center; (7) clustering number screening is carried out; (8) the best clustering number is selected; according to the method, the problems of high cost and low prediction accuracy of water quality prediction of small and medium-sized reservoirs are mainly solved. Meanwhile, the problem that a traditional clustering algorithm is inapplicable to data heterogeneity and differentvariances is solved, and the training accuracy of the NARX neural network is improved to a certain extent.

The invention relates to an automobile collision protection device, a monitoring protection method and an automobile equipped with a collision protection device, the device comprises an ultrasonic ranging and speed measuring module, control display module, voice alarm module, A lase fog lamp adjusting module and a buff airbag ejecting module, The ultrasonic ranging and speed measuring module usestwo ultrasonic groups before and after to calculate the relative distance, speed and acceleration information between the vehicle and the front and rear vehicles through the ultrasonic ranging methodof sending square wave and receiving square wave and combining the sound speed compensation of the temperature compensation module. The ultrasonic ranging and speed measuring module is used to calculate the relative distance, speed and acceleration information between the vehicle and the front and rear vehicles. The voice alarm module is used for voice real-time broadcasting; The laser fog lamp adjustment module is used to warn the rear vehicle of the occurrence of dangerous conditions; The cushioning airbag ejection module releases the airbag from the vehicle body when the system calculates and judges that the current driving condition of the vehicle is in a state where a collision can occur at any time, and rapidly inflates the airbag, thereby playing a cushioning role.

The invention belongs to the technical field of digital image processing, in particular to an image super-resolution reconstruction method driven by semantic segmentation. The method of the inventionspecifically comprises the following steps: separately training a super-resolution network model and a semantic segmentation network model of an image; cascading the independently trained super-resolution network and semantic segmentation network; driven by the task of semantic segmentation, training the super-resolution network. After the low-resolution image is processed by the task-driven network, the accurate semantic segmentation results are obtained. The experiment result shows that the invention can make the super-resolution network better adapt to the segmentation task, provide clear and high-resolution input images for the semantic segmentation network, and effectively improve the segmentation accuracy of the low-resolution image.

The invention discloses a license plate detection method based on end-to-end multi-task deep learning. The method is used for solving the technical problem that an existing license plate detection method is low in accuracy. According to the technical scheme, features of different hierarchies of an input picture are extracted through a full convolutional neural network, then two feature merging branches are used for regression to obtain a license plate and a bounding box of a corresponding vehicle, and finally license plate detection and vehicle detection are achieved at the same time. Due to the fact that convolution characteristics of different stages are utilized, license plate detection under the variable-scale condition can be well coped with. Through multi-task deep learning, high detection recall rate and accuracy are achieved, and high robustness is still achieved in a complex scene.

The invention relates to a dirt self-checking method of a digital camera image sensor based on retina perception, which comprises the following steps of: 1, placing a pure color object in front of a digital camera lens, and photographing to obtain image data of a frame of RGB channel; 2, reading image data and carrying out graying by a priori method; 3, setting a convolution kernel, and carrying out convolution-based mean value downsampling; 4, performing convolution calculation and retina perception-based self-quotient image calculation to obtain data in a floating point form; 5, performing data processing; 6, performing normalized cross correlation template matching by adopting a normalized cross correlation algorithm to obtain a correlation heat map; 7, performing thresholding processing on the correlation heat map; and 8, analyzing the connected domain to obtain the dirty position of the digital camera image sensor. The method can quickly and effectively extract and detect the dirtcharacteristics of the digital camera image sensor, and is low in detection cost, high in processing speed and easy to implement.

The invention provides an integrated estimation method for the health state of a retired power lithium battery, and the method comprises the following steps: firstly obtaining an aging feature for health state estimation from an applied combined current pulse test; secondly, using a multi-objective optimization method, optimizing the establishment method of the weak learner from two aspects of thenumber of aging features and the estimation precision; on the basis, using a selection method based on Pareto surface distribution characteristics to further reduce the number of weak learners, meanwhile, the precision of each weak learner is kept, and the overall efficiency of integrated estimation is improved; finally, obtaining each weak learner through training according to an optimal weak learner scheme, and realizing the integrated estimation of the charge state of the retired power lithium battery by adopting a differential evolution algorithm and combining a plurality of weak learnersin a weight mode. Through the design, the data can be actively analyzed, the optimal design of an ensemble learning scheme is automatically realized, and the accuracy and robustness of retired powerlithium battery health state estimation are improved.

A system for processing phone calls comprising a network of servers in communication with telephonic communication devices, and software stored on and executable by the servers, the software being configured to receive inputs from a user, generate an ordered set of PBX Feature definitions executable by the server network in a distributed fashion to perform respective PBX Feature tasks related to telephone calls, and write the PBX Feature definitions to a distributed database system that is accessible by the servers. The ordered set may evaluate the current state of a phone call, and each PBX feature definition may perform a respective task responsive to the evaluated state. Each PBX Feature definition is executable to define and store a new state of the phone call. Each PBX Feature definition generated by the software is configured to be combined with any other PBX Feature definition generable by the software.

The present invention is directed toward a system and process that controls a group of networked electronic components using a multimodal integration scheme in which inputs from a speech recognition subsystem, gesture recognition subsystem employing a wireless pointing device and pointing analysis subsystem also employing the pointing device, are combined to determine what component a user wants to control and what control action is desired. In this multimodal integration scheme, the desired action concerning an electronic component is decomposed into a command and a referent pair. The referent can be identified using the pointing device to identify the component by pointing at the component or an object associated with it, by using speech recognition, or both. The command may be specified by pressing a button on the pointing device, by a gesture performed with the pointing device, by a speech recognition event, or by any combination of these inputs.