244results about How to "Improve scene adaptability" patented technology

The invention provides an academic resource recommendation service system and method. The method comprises the following steps: crawling academic resources on an internet by using an LDA (Latent Dirichlet Allocation)-based focused crawler, classifying the academic resources according to preset A types by using an LDA-based text classification model, and storing the academic resources in a local academic resource database, wherein the system further comprises an academic resource model, a resource quality value calculation module and a user interest module; implanting a tracking software module at a user terminal, combining interesting subjects and historical browsing behavior data of the user, respectively modeling the academic resource model and the user interest module by virtue of four dimensions such as the academic resource type, subject theme distribution, key word distribution and LDA latent theme distribution, calculating the similarity between the academic resource model and the user interest preference module, combining the resource quality value to calculate the recommendation degree, and finally perform academic resource Top-N recommendation for the user according to the recommendation degree. According to the method disclosed by the invention, personalized accurate recommendation of the academic resources is performed according to the identity, interest and browsing behaviors of users, and the working efficiency of scientific research personnel is improved.

The invention provides an image information and radar information fusion method and system for a traffic scene, and the method comprises the following steps: preprocessing the image information, obtained by a camera, of the front of a vehicle; extracting the feature information in the image information, and comparing the feature information with the pre-stored traffic scene information; classifying the current traffic scene according to a comparison judgment result, performing a corresponding fusion algorithm according to a preset fusion method of image information and radar information whichis adapted to the current traffic scene category, and outputting the result of the fusion algorithm. The method judges a scene according to the collected image information, switches between differentfusion algorithms, effectively utilizes resources, and improves scene adaptability. The method fully utilizes redundancy and complementary characteristics between different sensor data to improve system robustness and reliability. The method adopts a deep learning algorithm in image information processing, which has higher real-time performance and is more accurate in target recognition.

The invention discloses a method and system for nighttime vehicle detection on the basis of vehicle light identification. The method includes: inhibiting a camera through strong light, obtaining traffic video data, and determining analyzed area-of-interest; performing filtering processing on images on the basis of mathematical morphology, and obtaining communication area information on the basis; then building vehicle light screening and matching rules, and screening and matching vehicle lights; performing position prediction on a target, setting vehicle confidence degree, updating stored vehicle position information, and achieving tracking of vehicles; and finally performing calculation and statistics on parameters of traffic flow, vehicle speed and the like of the vehicles left the area-of-interest. The method and system for the nighttime vehicle detection on the basis of the vehicle light identification adopts a large number of vehicle light image data statistic priori knowledge to identify vehicle light pairs and selects the vehicle light pair with highest matching degree from a connection area. The vehicle identification method enables the vehicle lights to be separated from nighttime scenes effectively and has good scene adaptability.

The invention discloses an extreme learning machine-based hyperspectral remote sensing image ground object classification method. An original extreme learning machine network is expanded into a hierarchical multi-channel fusion network. In terms of network training, the method is different from the least squares algorithm-based output weight solving strategy of the original ELM (extreme learning machine) and the global iterative optimization strategy of a deep learning network; according to the method of the invention, a greedy layer-by-layer training mode is adopted to train a hierarchical network layer by layer, and therefore, the training time of the network is greatly shortened; and in the layer-by-layer training process, a l1 regular optimization item is added into the training solving model of each layer of the network separately, so that parameter solving results are sparser, and the risk of over-fitting can be lowered. In terms of network functions, A single-hidden layer ELM network focus on solving the fitting and classification problems of simple data, while the different levels of the network model provided by the invention achieve target data feature learning or feature fusion, the network model of the invention integrates the advantages of high training speed and strong generalization capacity of the single-hidden layer ELM network, and therefore, the in-orbit realization of the model is facilitated, and the requirements of emergency response tasks can be satisfied.

The invention discloses a moving object relay tracing algorithm of multiple PTZ (pan / tilt / zoom) cameras. The moving object relay tracing algorithm is characterized in that the moving object relay tracing algorithm is carried out according to the following steps: 1. estimating inner parameter matrixes of the PTZ cameras by adopting a camera self-calibration method; 2. setting a view cutting line between adjacent PTZ cameras; 3. using Logistics regression model as a classification function, and combining with a mean value drifting algorithm and realizing the target tracking; 4. continuously adjusting angles of the PTZ cameras in the course of tracking, so that the target is always located in the center zone of the PTZ camera view; 5. when the target exceeds the view cutting line of the PTZ camera at present and enters the monitoring view of the adjacent PTZ camera, calculating the coordinate of the target in the adjacent PTZ camera view, and transferring the adjacent PTZ cameras to track the target, and rotating the original PTZ camera to return the preset position. The moving object relay tracing algorithm can accurately control the camera rotation, and perform stable tracking for a long time on the target; and thereby, the complete historical movement information of the target can be obtained.

The invention relates to a multi-source information fusion positioning method and device. The multi-source information fusion positioning method comprises the steps: (1) the track is deduced through asensor integrated in an intelligent terminal; (2) weighted centroid positioning and signal sequence matching positioning are conducted through a Bluetooth sensor; (3) weighted centroid positioning and signal sequence matching positioning are conducted through a Wi-Fi sensor; and (4) the track deduction result obtained in the step (1) and the positioning results obtained in the step (2) and the step (3) are fused, and the final positioning result is obtained. An inertial navigation / Bluetooth / Wi-Fi-based fusion positioning mode is adopted, the advantages of various positioning technologies canbe developed effectively, and high-precision positioning and navigation are achieved; and the positioning cost is low, deployment is convenient, the coverage range is wide, the very high scene adaptability is achieved, real-time, continuous and accurate position information can be provided, and the multi-source information fusion positioning method and device can be widely applied to various indoor positioning scenes.

The invention discloses a fusion parking system, which comprises a camera, an ultrasonic sensor and a fusion controller, wherein the camera, the ultrasonic sensor and the fusion controller are arranged on a vehicle. The camera acquires camera 3D point cloud information of an obstacle and a parking space in the traveling process of the vehicle through an image mode, the ultrasonic sensor acquires the ultrasonic 2D point cloud information of the obstacle in the traveling process of the vehicle through an echo mode, the parking space information detected by the camera and the ultrasonic sensor isdivided into a non-obstacle and an obstacle, and a distance measurement precision training of the camera and the ultrasonic sensor is carried out on various kinds of information so as to obtain different groups of confidence levels. The sensor information is fused according to different groups of confidence degrees of the sensor, so that an obstacle parking space and a line-marked parking space can be accurately identified.

The invention belongs to the technical field of image processing, and provides a target tracking method and device, a computer device and a storage medium. The target tracking method comprises the following steps: detecting a predetermined type of target in a current image by using a target detector to obtain a first target box in the current image; Obtaining a second target frame in the previousframe of image of the current image, and predicting the position of the second target frame in the current image by using a predictor to obtain a prediction frame of the second target frame in the current image; Matching a first target frame in the current image with the prediction frame to obtain a matching result of the first target frame and the prediction frame; And updating the position of atarget in the current image according to a matching result of the first target frame and the prediction frame. According to the invention, the robustness and scene adaptability of target tracking areimproved.

The invention discloses a target tracking method based on local feature learning. A target object and background are resolved into a large number of local units having dimension and shape invariance which serve as training samples of a target and background classification model, and a mode of deep learning is adopted to learn local expression of the target object and background from the training samples. The degree of confidence that each specific area in an image belongs to the target object is judged, thereby realizing accurate positioning of the target object. Since local expression obtained by learning of a large number of samples has a high target identification capability, the tracking method has relatively high adaptability to situations such as target deformation and target occlusion. When an object appearance model is updated, local areas with a relatively high target confidence degree are extracted as training samples of the model, and the model is updated. In the process of target tracking, the method continuously updates the appearance model, learns key characteristics of the target object, and can obtain a relatively good tracking effect under the circumstance that an appearance change is relatively big.

The invention discloses an MEC server selection algorithm based on fairness, and belongs to the field of mobile communication. The algorithm comprises the steps of establishing a system model among users, channels and MEC servers; randomly selecting the users to be connected with the MEC servers by an initial system; sending calculation tasks in time t by the selected users, and matching the MEC servers for the users according to communication cost; uploading the calculation tasks by the users, and calculating the throughput of each user; sorting the throughput of all users in a descending mode, selecting former N users with the low throughput, and calculating the communication cost of the users; and for the users with the low throughput, selecting the MEC serves by the users for next time calculation. The algorithm has the advantages that the influences of delay and energy consumption on selection of the users to the MEC servers are taken into combined consideration, and a communication model is introduced into the system; and when the users are in different application scenes, different weight parameters are selected, the communication cost is controlled, and the algorithm has better scene adaptability.

The invention discloses a specific scene model upgrading method based on federated learning. The method comprises the following steps: selecting an AI terminal; marking all scene data in a marking module of each AI terminal, storing the scene data in the local of the terminal, calling an intelligent classification and statistics module to perform classification statistics on the scene data, and reporting the scene data to an alliance statistics module; forming a federated learning alliance by the terminals similar to the usage scenarios, wherein after a preset updating condition is met, the alliance statistical management module triggers and starts transverse federation learning training of the updating model; screening the training data before the training of the updating model; performing a learning updating process by the transverse federation; in the process, the preset updating condition can be modified along with the increase of the number of terminals and the increase of data volume, and iterative updating is continued. According to the invention, the problems of low model updating and upgrading efficiency of the artificial intelligence terminal and low accuracy in an actualuse scene can be solved.

The invention relates to a movable illegal parking snapshot system and an implementation method. The implementation method includes the steps: parking a movable detection automobile at a detection position; lifting a snapshot unit to a required height; configuring basic parameters and communication parameters of the snapshot unit and an industrial personal computer; drawing an initial model of a detection area, setting illegal parking judgment rules, and drawing adjusting points of the area; adjusting six adjusting points of the area to obtain a detection area A when a detection road section is a linear road section; adjusting six adjusting points of the area to obtain a detection area B when a detection road section is a bent road section, and performing snapshot when an illegal parking behavior is established as total parking time of a vehicle target in an illegal parking area exceeds a time threshold value preset by a system; transmitting illegal parking data information to a command center platform by a transmission device. The movable illegal parking snapshot system performs snapshot for a large number of illegal parking behaviors of fixed illegal parking detection snapshot monitoring ranges, control ranges of roads are widened, control time of the roads is prolonged, law-abiding consciousness of drivers is improved, and road traffic orders can be greatly improved.

The invention belongs to the technical field of image processing, and particularly relates to a multi-point correction method for an infrared focal plane. The method comprises steps as follows: step 1, an infrared focal plane detector acquires images of different temperature black bodies at the working temperature in every other 5 DEG C and uses the images as background images; step 2, a shutter is started, a related parameter of a shutter image is acquired, and a shutter correction parameter is acquired with the following formula shown in the specification according to the background images, wherein delta O is the shutter correction parameter, the related parameter of the shutter image comprises the mean value Xs of a shutter image Xs and the shutter image, Bi is the i background image, Bi is the mean value of the background images, Ks is a gain correction parameter corresponding to the shutter image, and Ks is the mean value of the Ks; step 3, according to the shutter correction parameter delta O, a gain correction parameter K and a bias correction parameter Bj which correspond to an object image are introduced into the following formula shown in the specification for two-point correction, and a corrected image Y is calculated, wherein X is the to-be-corrected image, Y is the corrected image, delta O is the shutter correction parameter, Bj is the j background image, and K is the gain correction image.

The invention provides a collision distance calculation method and system based on fusion of an ultrasound wave and a camera. On the basis of obstacle detection based on ultrasound and camera fusion,different types of obstacles such as vehicles and pedestrians, pure ultrasound waves are obtained; collision distances are calculated respectively according to vehicle parameters, turning radiuses, vehicle speeds, and different types of obstacle information; areas of interest are selected in different driving directions and turning directions; and distance of all obstacles in the areas of interest are minimized to obtain the fused collision distances. According to the invention, with the existing ultrasonic sensor, all-around CMOS camera, integrated parking controller and all-around controller in a current vehicle, the relative position accuracy of detection is improved and the obstacle detection reliability and detection distances are enhanced. The collision distance calculation methodand system have advantages of reliable detection, high accuracy, high scene adaptability, and moderate cost; and the adaptability of the low-speed auxiliary system is improved effectively.

The invention relates to a video perceived quality non-reference objective evaluation method based on areas of interest. The method is technically characterized in that the areas of interest including motion areas and human skin areas are extracted from a video to be detected, the two areas of interest are fused, the final area of interest is obtained, and different evaluation weights are given to the area of interest / non-interest areas; image distortion degrees are evaluated frame by frame; the evaluation weight of each image and two evaluation results are combined and switched so as to be matched with a subjective evaluation result. The method is reasonable in design, the motion areas and the human skin areas are comprehensively considered and blocking effects and fuzzy distortion characteristics in the two images are combined, video quality can be evaluated in a non-reference mode, good scene adaptability is achieved, the good correlation between the evaluation results and the subjective evaluation result is shown, and the method can be used for video quality evaluation and image quality evaluation.

The invention provides a nonlinear-fitting infrared non-uniform correction method based on time-domain Kalman filtering correction. The technical proposal is as follows: firstly, assuming that a response change curve of each array unit of an infrared focal plane detector is continuous on time, and using a high-order multi-parameter nonlinear polynomial to fit and describe the response change curve; secondly, acquiring a response output value of each array unit at four different temperature points, using a nonlinear equation solution method to determine a plurality of parameters in each array unit response expression; thirdly, using time-domain Kalmn filtering to carry out time-shift correction on each parameter in the response expression for solving a problem that the detector response shifts with the time, and acquiring an analysis expression of the response curves of the detector array unit; and finally, using the analysis expression to calculate the response output of each array unit of the detector under any time and any temperature conditions. The nonlinear-fitting infrared non-uniform correction method realizes the correction of inconsistent response of the detector and solves non-uniformity problem of the infrared image.

The invention provides a mobile terminal and a display control method thereof. The mobile terminal comprises a main board, a display screen which is connected with the main board, a data acquisition unit which is connected with the main board and is used for acquiring motion computational data, a computing unit which is connected with the main board and is used for computing motion information of the terminal through the motion computational data, a first control unit which is connected with the main board and is used for controlling a display content selection unit displayed on the display screen according to the motion information so as to select display contents displayed on the display screen as first display parameters, and a second control unit which is connected with the main board and is used for controlling the selected display contents to be displayed on the display screen as second display parameters, wherein the second display parameters are different from the first display parameters. The mobile terminal and the display control method thereof have better scene adaptability.

This invention provides a global motion vector estimating method and a monitoring video stabilizing method. A final global motion vector is computed according to a global motion vector of an upper sampling image and a corrected global motion vector; feature points are detected and matched in a lower sampling image to quickly estimate a global motion; and then, block matching is performed on the matched feature points; the block matching has higher accuracy and efficiency. A jitter motion vector is computed according to the difference between a global motion smooth vector and the global motion vector; whether the jitter motion vector is out of a preset range or not is judged; if not, an image compensation is performed according to the jitter motion vector to output video stabilizing images. By adopting the methods provided by the invention, aiming at the characteristics of the monitoring video, the images are stabilized; the technical selection is more reasonable and specific; the algorithm with low complexity is selected on the premise of ensuring an effective algorithm; and thus, the methods are more efficient in real time.

The invention provides an image enhancement method for scene self-adaptive wide dynamic infrared thermal imaging, which belongs to the technical field of image processing. The method comprises the following steps of calculating to obtain an initial low-frequency base layer image by using a bilateral filtering algorithm according to a 16-bit original image; calculating to obtain an initial high-frequency detail image; performing adaptive histogram equalization (CLAHE) operation on the initial low-frequency base image to obtain a first 8-bit low-frequency base image; performing global histogramequalization processing on the initial low-frequency base image to obtain a second 8-bit low-frequency base image; obtaining a final high-frequency detail image by utilizing automatic gain control operation according to the initial high-frequency detail image; utilizing linear weighting calculation to obtain a final low-frequency base layer image; and fusing the final high-frequency detail image and the final low-frequency base image to obtain an enhanced output image. According to the method, problems of poor scene adaptability and over-enhancement of the existing infrared thermal imaging wide dynamic range image enhancement technology are solved.

The invention discloses an image-based crowd counting method and image-based crowd counting equipment. The image-based crowd counting method comprises the steps of inputting an image to be analyzed into a neural network comprising a multi-layer convolution layer and a feature fusion layer, performing scale normalization processing on output features of feature extraction sub-networks by means of the feature fusion layer, combining the output features after scale normalization processing with weighting coefficients corresponding to the output features to generate a crowd density map of the image to be analyzed, and finally integrating the crowd density map to determine the number of people in the image to be analyzed. The image-based crowd counting method and the image-based crowd countingequipment can count the total number of people based on one single image, have the advantages of high accuracy rate and good adaptability to the scene, can simultaneously satisfy dense crowd countingand sparse crowd counting, and can greatly reduce the workload of the statistician.

The invention discloses a vehicle image optimization method and system based on adversarial learning. The vehicle image optimization method comprises the steps: collecting vehicle images photographedat different angles, and dividing the vehicle images into a standard scene image and a non-standard scene image; carrying out image preprocessing on the non-standard image to obtain a low-quality dataset; constructing a vehicle image optimization model based on the generative adversarial network, wherein the model is composed of a generator, a discriminator and a feature extractor; training a vehicle image optimization model based on the generative adversarial network, setting a loss function, calculating a network weight gradient by adopting back propagation, and updating parameters of the vehicle image optimization model; and after the vehicle image optimization model is trained, reserving the generator as a final vehicle image optimization model, inputting multi-scene vehicle images, and outputting optimized standard scene images. According to the invention, migration from complex scene vehicle images to standard scene vehicle images is realized, and the purpose of optimizing the image quality is achieved, and the vehicle detection and recognition accuracy is improved.

The invention discloses a mapping method of a file system access control list and is applied to file server. An operation request of a client for a target node in the file system can be responded; anaccess control list of the client can be obtained, whether the format of the access control list of the client is the same as that of the access control list of the client or not is judged; and if not, the format of the access control list of the client is converted into the format of the access control list of the client to obtain a target access control list, so as to judge whether the client has permission to execute the operation request by inquiring the target access control list. Therefore, according to the method, when the client initiates the operation request, the format of the access control list is unified, the problem that access cannot be achieved is avoided, and the scene adaptability of the file server is improved. In addition, the invention further provides a mapping device of the file system access control list, a file server and a storage system, and the effects of the mapping device, the file server and the storage system correspond to those of the method.

The invention discloses a gray scale and gradient segmentation-based infrared target detection method. The method comprises the following steps: 1) calculating a gradient image Phi(x, y) by utilizing an original image Forg; 2) carrying out non-maximum value inhibition on the Phi(x, y) to obtain an image Phi1(x, y); 3) carrying out edge linking and edge closing judgement by utilizing the Phi1(x, y), and calculating a candidate target area Pos1; 4) selecting a local image Floc according to the Pos1 and segmenting the Floc to obtain a candidate target area Pos2; and 5) calculating an optimum target area Pos according to a formula (as shown in the specification).

The invention provides a visual detection method for pantograph-catenary arcing of an electrified railway, and relates to the technical field of computer pattern recognition. Firstly, mask labeling is carried out on a pantograph-catenary image with arcing, then a labeled image is used as a data set of a multi-dimensional feature fusion segmentation network to train a network, the segmentation network adopts a deep convolutional network and is composed of a feature extraction module, a multi-dimensional feature fusion module and a head module, forward reasoning is carried out on the pantograph-catenary image through the segmentation network, and a network segmentation head sub-module outputs a feature map obtained after double up-sampling of a result as a segmentation result of the pantograph-catenary image. Deep separable convolution and grouping convolution are added into a multi-dimensional feature fusion module, the network is enabled to pay more attention to arcing region features through the addition of a same-channel attention and space attention mechanism, the network can accurately detect whether an arcing phenomenon occurs in a pantograph-catenary image after training is completed, and the accuracy and robustness of the network can also be improved by performing online learning and adaptive switching on the model.

The invention relates to a fish body posture and length automatic analysis method based on key point detection and a deep convolutional neural network, and the method comprises the steps: S1, obtaining binocular images comprising a fish school through an underwater binocular camera, wherein the binocular images comprise a left image and a right image; and S2, performing calibrating in an underwater environment to obtain binocular camera parameters, and performing binocular correction on the obtained binocular image. The beneficial effects of the invention are that the method combines the deepconvolution neural network, and is high in adaptability to an application environment and a scene; a key point detection idea is introduced, and only the spatial positions of specific key points on fish bodies are concerned, so that the difficulty of global binocular matching in underwater application is avoided; required equipment is simple, and only an underwater binocular camera and an operation rear end are required; attitude estimation and length measurement can be carried out on multiple fishes with different positions and attitudes in the image in real time, and the accuracy and the efficiency are relatively high; and the model also has generalization ability for tasks, and is easy to migrate from one working scene to another.

The invention discloses image capture equipment and a scanning target extraction method and extraction device thereof and a storage medium. The scanning target extraction method comprises the steps that a scanning area image is acquired; the closed contour lines of all the objects in the scanning area image are extracted, wherein the objects include a target object; the first angular points of the closed contour lines are recognized according to the first preset rules; whether the number of the first angular points is consistent with that of the preset angular points is judged; if the judgment result is yes, the image of the target object is extracted from the scanning area image according to the first angular points; and if the judgment result is no, the second angular points of which the number is consistent with that of the preset angular points are determined according to the closed contour lines and the preset algorithm, and the image of the target object is extracted from the scanning area image according to the second angular points. With application of the method, the target object can be effectively extracted from the scanning area image when the target object is arbitrarily shaped or the interference object exists so that the scene adaptability of the image capture equipment for scanning can be enhanced.

The invention provides a reminding system and a reminding method for an intelligent door lock. The reminding method comprises the following steps: detecting a door closing signal through a first sensor, detecting a locking signal through a second sensor and mentoring a signal that a user leaves the door lock through a non-contact sensor; and in the door closing process, if the door closing signaland the information that the user leaves the door lock are detected but the locking signal is not detected, sending out reminding information. The reminding system judges that the user leaves the doorlock by way of a non-contact detection mode with the door lock, is more flexible in reminding control and high in scene adaptability, and is particularly suitable for the user carrying things to go out of the door and come into the door.

The invention discloses a solid state disk life prediction method, device and apparatus and a readable storage medium. According to the scheme, a medium abrasion index and a timestamp of a solid statedisk within the preset time range are obtained by utilizing physical characteristics that programming and erasing cycle index of the solid state disk is fixed. A prophet model is constructed based on the medium abrasion index and the timestamp so as to predict the service life of the solid state disk. The prophet model can detect the trend change, periodic change and holiday change of the wear degree of the solid state disk. Compared with a simple linear formula, scene adaptability is stronger, so that service life of the solid state disk can be accurately predicted, a user is assisted to know the service condition of the solid state disk, loss caused by damage of the solid state disk is avoided, and safety of data in the solid state disk is remarkably improved.

The invention discloses an automatic focusing method and apparatus. The method comprises: multi-frame image data are obtained continuously and focusing window division is carried out on each frame of image data among the multi-frame image data; grey level transformation processing and filtering processing are carried out on the image data in each focusing window successively; for each pixel point in the image data after filtering processing, a first gradient value and a second gradient value that correspond to the pixel point are calculated; according to the first gradient values and second gradient values corresponding to all pixel points in each focusing window, a first focusing evolution value and a second focusing evolution value that correspond to each focusing window are obtained; and a focusing direction of a motor is determined according to the first focusing evolution values and second focusing evolution values and a step length of the motor is adjusted based on a first focusing changing rate of each first focusing evaluation value and a second focusing changing rate of each second focusing evaluation value until the lens of the motor is adjusted to be at a focusing state. Therefore, the focusing speed is fast and accurate.

The invention discloses a method of dynamically ranging a target with a vehicular single camera, comprising steps of sampling a 0-momnet picture in video images of a camera, defining it as a first-frame picture PT [0], graying the picture, recognizing a target in the grayed picture to obtain a typical target outline, marking with a square, and recording image height PTH(0) ad image width of the square image of the target at this moment; sequentially recording PT[1], PT[2] to PT[i] images of adjacent i frames to obtain image height PTH(N) and image width PTW(N) of an image of the target in the PT[N] image; recording current vehicle speed; calculating by deducting to obtain a distance from the target. According to the method, the single camera is used with no assistance; the method has good timeliness, an algorithm is simple, and the method is easy to combine with videos in real time; the method is independent of optical parameters and target object characteristics; the method is highly adaptive to scenes and is low in cost and high in practicality.