94results about How to "Achieve real-time" patented technology

The invention discloses a small target detection method based on feature fusion and depth learning, which solves the problems of poor detection accuracy and real-time performance for small targets. The implementation scheme is as follows: extracting high-resolution feature map through deeper and better network model of ResNet 101; extracting Five successively reduced low resolution feature maps from the auxiliary convolution layer to expand the scale of feature maps. Obtaining The multi-scale feature map by the feature pyramid network. In the structure of feature pyramid network, adopting deconvolution to fuse the feature map information of high-level semantic layer and the feature map information of shallow layer; performing Target prediction using feature maps with different scales and fusion characteristics; adopting A non-maximum value to suppress the scores of multiple predicted borders and categories, so as to obtain the border position and category information of the final target. The invention has the advantages of ensuring high precision of small target detection under the requirement of ensuring real-time detection, can quickly and accurately detect small targets in images, and can be used for real-time detection of targets in aerial photographs of unmanned aerial vehicles.

The invention discloses a real-time face optimal selection method based on a video sequence. The method comprises the following steps: a face image is acquired; face clarity, face size and opening degree of human eyes act as three indexes for face quality evaluation so that a face integrated evaluation score ImageScore is calculated; the ImageScore of the first frame of the face image is assigned to an initial face optimal image MaxScore and the face image is stored; the MaxScore after assignment is compared with the ImageScore of other frames of the face image, and the face image corresponding to the higher ImageScore is stored and the ImageScore is assigned to the MaxScore again; and the MaxScore value is cyclically updated and the stored optimal face image is updated until the image sequence with the face cannot be acquired, and the cycle is ended and the image corresponding to the stored MaxScore is the optimal face image. The face image with higher quality can be screened for identification via comparison of the integrated evaluation indexes of face quality in the video sequence under the situation of not knowing the face sequence image in advance; besides, the calculation method is simple with real-time performance.

The invention discloses a small target detection method based on regional nomination. The method comprises: step one, carrying out input image preprocessing and converting the input image into one with a unified size; step two, extracting an image base feature map and extracting five-layer feature maps; step three, carrying out feature fusion and carrying out fusion of the fourth-layer feature mapand the fifth-layer feature map; step four, carrying out small target area nomination and generating small target area nomination by using a regional nomination network; and step five, carrying out small target bounding frame refinement and small target area classification. The method has advantages of fast calculation speed, high small target recognition precision, good generalization performance, and good detection performance of common minimum target areas.

The invention provides a real-time human body detection method based on an AdaBoost frame and colour of a head, comprising training and detecting steps: (1) extracting multi-scale HOG (histograms of oriented gradients) characteristics according to a template; (2) training a human body detection model by adopting an AdaBoost-Boosting method; (3) extracting histogram characteristics of the colour of the head; (4) training a head discrimination model by adopting an AdaBoost method; (5) detecting a human body based on a sliding window method; (6) for each detection window, firstly extracting the HOG characteristics, and judging whether the detected object is a human body or not according the human body detection model; and (7) for the window which is determined to be a human body, extracting the histogram characteristics of the head, judging whether the head is contained or not; and determining the window containing the head to be a window containing a human body, and drawing a rectangle at a corresponding position in an image. In the invention, the characteristic unit, namely a Block of the original HOG characteristics is adopted, the Block is of multi-scale type, and a characteristic template is combined, thus the detection effect is improved; and the head characteristic discrimination is added, thus the detection rate is improved.

A congestion prediction algorithm based on historical and peripheral intersection data comprises a first step of performing real-time updating and preprocessing on traffic data; a second step of inputting a predicted target intersection and a time interval; a third step of matching the traffic flow data with historical data of the current area to find out a similar time interval and then carryingout prediction. According to the method, the problem of " traffic flow prediction " is solved by utilizing an algorithm idea of non-parameter regression, the traffic flow is predicted according to historical data and peripheral intersection data, and the traffic data flow is effectively predicted. The congestion situation of the target intersection at the demand time is predicted according to therequirement of a user on the traffic jam prediction situation, and the traffic flow can be efficiently and accurately predicted.

The invention discloses a model construction method, and particularly relates to a face image real-time beautifying and texture synthesizing method for three-dimensional face reconstruction. Accordingto the method, through the processes of face feature point extraction by utilizing an ASM algorithm, color conversion, face seamless fusion, eyebrow repair processing, three-dimensional face reconstruction and the like, the automatic detection and beautification can be realized on the face, so that a relatively better face reconstruction result is obtained. Experimental results show that the algorithm provided by the invention can obtain a satisfactory reconstruction effect, including reconstructing a shielded eyebrow region, has the superior operation efficiency compared with other methods,and can achieve the real-time performance. The face image is beautified based on the standard skin color to generate the face texture map, and the face texture map is applied to the three-dimensionalface reconstruction to obtain the personalized three-dimensional face model.

The invention discloses a real-time detection system for wood microstructure characteristic changes, which comprises a loading unit, a microscopic unit, an image acquisition unit and an image processing unit, wherein the loading unit is used to stretch, compress, bend, twist or cut to-be-detected wood; the microscopic unit is used to magnify a to-be-detected area of the wood; the image acquisition unit is used to acquire surface image information of the area magnified through the microscopic unit; and the image processing unit is used to receive image information acquired by the image acquisition unit, calculate and obtain the micro-area strain information and the real-time microstructure change results of the to-be-detected wood. The system can measure micro-area tissue and cell strain to realize micro-area deformation quantization, and can monitor micro-area defects.

The invention discloses a method and device for processing stock data of medical consumables. The method comprises the following steps of: storing the stock data of the medical consumables, wherein the stock data comprise stock quantity; judging whether the stock quantity is smaller than a preset threshold value; and outputting a prompt signal under the condition that the stock quantity is smaller than the preset threshold value. By means of the method and the device, real-time monitoring and early warning can be performed on delivery conditions of the stocked consumables.

The invention discloses a corn breakage rate detection method based on a GPU embedded platform. The method comprises three portions. A first portion is classification model generation. A second portion is image segmentation, which contains that adhesive corn grain images are segmented so as to extract individual grain images and parallel processing is performed on a segmentation algorithm. A third portion is that a classification model is used to carry out breakage detection on the segmented images and a breakage rate is calculated. In the invention, convex set limit corrosion and condition expansion algorithms are used to effectively solve a segmentation problem of the adhesive corn grain images; a deep learning method is used to avoid a manual characteristic extraction problem and guarantee breakage rate detection accuracy; and the GPU embedded platform is used to carry out parallel processing on a detection algorithm so that the detection algorithm possesses real-time performance.

The invention discloses a human flow statistics method based on RGB and D information fusion, which comprises the following steps: step 1, collecting an image sample containing human head information;2, marking that head area of the RGB sample, cutting the positive and negative sample, regularizing the size of the cut sample image and attaching corresponding sample labels to different samples; 3,extracting HOG features of all RGB sample image, and inputting that extracted HOG features and the corresponding sample tag into an SVM classifier for training; 4, detecting the head area of the RGBimage in the newly inputted video frame by using the trained SVM classifier; 5, detecting the depth information of Depth image of the newly inputted video frame; 6, fusing that detection results to obtain a final detection target frame; 7, tracking the target object to obtain the moving trajectory of the target object, and counting the human flow information.

The invention discloses a configurable image data caching system based on an FPGA and a DDR3SDRAM. The system comprises an external storage module, a storage control module, a first cache module, a second cache module and a cache control module, and video image data and clock signals are written into the first cache module from the cache control module for first-stage caching; the storage controlmodule controls to read the data from the first cache module and store the data in an external storage module; then the storage control module controls to read the stored data from the external storage module, and writes the data into the second cache module for second-stage caching; then, the cache control module controls to read the data from the second cache module to the cache control module,the cache control module adjusts the image data and generates corresponding read effective signals, and the configuration data are output to the outside; according to the invention, larger data transmission can be cached, and the effects of real-time performance and stability of data transmission are achieved; and the packaged system only has a few ports, so that the configuration is more convenient.

The invention discloses an oil and gas pipeline corrosion and leakage monitoring and early warning system and method based on OFDR optical fiber sensing. The system comprises distributed optical fiberstrain sensors, distributed optical fiber temperature sensors, a pipeline internal pressure sensor, an OFDR optical fiber data automatic acquisition device, a pavement base station, a pavement base station information display and transmission system, a solar power supply system, a 5G signal transmission system, an optical fiber data processing and analysis system and a display early warning system. The strain sensors annularly laid on the surface of a pipeline are used for monitoring pipeline corrosion through hoop strain, and the pipeline leakage is monitored by the linear laid optical fibertemperature sensors. After the OFDR data acquisition device sends data to the pavement base station information display and transmission system, the hoop strain data or the pipeline surface temperature is transmitted to the optical fiber data processing and analysis system through the 5G transmission system, and the display early warning system monitors and conducts early warning after comparingthe result with a preset alarm value. According to the system, the degree and the size of corrosion are determined through data comparison, the change of the temperature of the outer wall of the pipeline caused by leakage of the oil and gas pipeline is monitored, and remote and site simultaneous monitoring is realized through the pavement base station.

The embodiment of the invention discloses a detection system based on Raspberry Pi. The detection system comprises the Raspberry Pi, a camera and a nerve calculation rod, wherein the camera and the nerve calculation rod are respectively connected with the raspberry pi; the camera is used for capturing image data; the Raspberry Pi is used for triggering the neural computing rod to acquire image data when acquiring the image data captured by the camera; and the neural calculation rod is used for inputting the acquired image data into a preset target recognition model to obtain an image recognition result output by the target recognition model, and sending the image recognition result to the raspberry pi. By applying the scheme provided by the embodiment of the invention, on the basis of realizing marginalized computing power, the identification efficiency can be improved so as to achieve real-time performance.

The invention provides a sea fog detecting method based on deep learning and satellite remote sensing technology. The sea fog detecting method based on the deep learning and satellite remote sensing technology comprises the following steps: obtaining a satellite remote sensing image, marking the sea fog in the image, and using the image marked with the sea fog as a label of a deep learning segmentation network model; pre-processing a training set image to obtain an input image that meets a preset criteria, wherein the input image is used for augmenting the data set and enhancing the model robustness; and using a back propagation algorithm to the deep learning segmentation network model on a GPU based on the augmented data set, so that a model of the image characteristics of the sea fog isobtained after the training is completed. The invention combines the deep learning with the satellite remote sensing to solve the problem that the sea fog detection is difficult. By learning the characteristics of the sea fog generated in coastal areas and oceans in a large number of past satellite images, the sea fog can be quickly and accurately monitored and segmented based on the satellite images.

The invention provides a method for detecting a yaw angle between a tractor and a trailer. The method comprises the following steps that, a detection board is arranged in front of a trailer, a plurality characteristic patterns are arranged in the front face of the detection board, and a camera is arranged in the rear face of the tractor and can be used for shooting at least one characteristic pattern on the detection board at an arbitrary yaw angle; under the situation of no yaw between the trailer and the tractor, and the camera is used for shooting the detection board to calculate an initialtransformation matrix; the characteristic patterns on the detection board are shot by the camera in a vehicle driving process; at least four characteristic points are extracted from each shot characteristic pattern; the current transformation matrix is calculated; and the yaw angle between the tractor and the trailer is calculated. The method is suitable for types of various trailers, including full trailers and semitrailers, has relatively low calculation strength, can realize real time, and can run under an extreme work condition (low light condition).

The invention discloses a real-time measurement method for wood deformation microstructure characteristics. The method comprises the following steps: 1) loading to-be-detected wood; 2) magnifying a to-be-detected area through a microscopic unit in the loading process; 3) acquiring image information through an image acquisition unit and then working out the X-axis length X0-Xn and the Y-axis length Y0-Yn of every cell on a loading surface at each image acquisition moment through an image processing unit; 4) calculating the pressure or tensile strain Epsilonxn and Epsilonyn of single cells on an X axis and a Y axis according to the X-axis length X0 and the Y-axis length Y0 of each cell before deformation, as well as the X-axis length Xn and the Y-axis length Yn of each cell after deformation, wherein Epsilonxn = (Xn-X0) / Xn, and Epsilonyn = (Yn-Y0) / Yn; and 5) averaging the pressure and tensile strain Epsilonxn and Epsilonyn of all single cells in the to-be-detected area magnified through the microscopic unit so as to obtain the pressure and tensile strain Epsilonx and Epsilony of a whole characterized wood on the X axis and the Y axis. The method can measure micro-area tissue and cell strain to realize micro-area deformation quantization, and can monitor micro-area defects.

The invention provides a real-time drawing method based on nonuniform sampling three-dimensional object reflection. The method comprises the following steps of acquiring a sampling templete of a reflected object image, generating a reflected object nonuniform sampling deformation pattern according to the sampling templete, and finally, drawing the reflection effect in the scene by the deformation pattern reconstructed nonuniform sampling image. The algorithm for automatically generating the sampling templete can draw the reflection phenomenon in real time. The reflection in the scene is more realistic when being drawn under the condition of low overall resolution (small reflection calculated amount), thereby achieving the goal of enhancing sense of reality.

The invention discloses a tire pressure monitoring system and a tire pressure monitoring device. The tire pressure monitoring system comprises the tire pressure monitoring device and a data acquiring device, wherein the tire pressure monitoring device is used for monitoring a tire parameter in real time, recording primary state data into an internal memory of the tire pressure monitoring device and sending the state data to the data acquiring device when the tire parameter reaches the preset value; the data acquiring device is used for acquiring the state data including recording time and tire parameters. According to the tire pressure monitoring system and the tire pressure monitoring device, the tire parameters including the pressure and the like can be monitored in real time, thus achieving real-time monitoring and pre-warning purposes; a driving state of an automobile in a certain time and the situation changes of the parameters including the pressure, the temperature, the accelerated speed and the like can be recorded, so that the track record to the driving state of the automobile is realized, and the stability and the reliability are improved; meanwhile, the system and the device can further play helping roles in the driving state analysis, troubleshooting and the like of the automobile after an event and even play certain reference roles in the accident cause affirmation of vehicles.

An on-line measuring system of pull type in deep sea is prepared as using transducer detection subsystem to detect deep sea data and transmitting detected data to hyper baric chamber measuring and controlling subsystem under water as well as transmitting said data to deck control subsystem after it is stored in hyper baric chamber measuring and controlling subsystem underwater, storing and processing as well as displaying said data by deck control subsystem.

The invention discloses a human body hidden thing image segmentation method based on an X-ray back scattering image. The method comprises the steps of collecting the X-ray back scattering image, preprocessing the X-ray back scattering image, obtaining an image after preprocessing, carrying out statistics on the preprocessed image, extracting human body areas, gaining the average grey degree u0 of the human body areas, classifying the human body areas again through a K-mean algorithm and a Bayesian classification algorithm and extracting human body hidden things by using a zone marker method. According to the human body hidden thing image segmentation method based on the X-ray back scattering image, the data size is reduced to the greatest extent under the condition that an effect can not be affected, so that computing time is shortened, and the purpose of real-time performance is achieved. Objects which have the relatively low contrast ratio with the background can be well detected. Through two-time segmentation, the purposes that the objects and the background are distinguished and the hidden things are well detected are achieved by using the middle background of a human body, the hidden things are automatically marked, and manual marking is avoided.

The invention discloses a lane line detecting method and device. The lane line detecting method provided by the invention has the advantages that a region of interest in a road image is acquired, wherein the region of interest is an image region including a lane line; the region of interest is equally divided into at least two sub regions; a straight line segment included in each sub region in atleast two sub regions is detected in parallel; detected straight line segments in each sub region are combined; and a lane line threshold is used to filter combined straight lines to acquire the lanelines. According to the invention, consumed time in a lane line detecting process can be reduced, and the real-time performance of the algorithm is improved.

The invention discloses a pig farm monitoring alarm system. The system comprises a video collection module, a query display module, a sensor module, a performance module, a controller module, an alarm module, a location module, a wireless module, a user terminal module and a data center module. The output ends of the video collection module, the query display module and the sensor module are connected with the input end of the controller module. The output end of the controller module is connected with the input ends of the performance module and the alarm module. The wireless module is connected with the controller module. The user terminal module is connected with the wireless module. The data center module is connected with the controller module. According to the system, the problems that an existing pig farm is difficult to ensure environment data collection timeliness and a great deal of manpower resources are wasted are solved.

The invention discloses a medical document identification method based on an LSTM neural network, and the method comprises the steps: 1) carrying out the preprocessing of a document image, and converting an image signal into a digital signal; 2) segmenting characters, normalizingf document images; 3) extracting character features, generating feature vectors; 4) performing document identification and classification.According to the method, an LSTM neural network is creatively adopted to recognize and classify the images, and the method has the advantages of being high in recognition speed, highin fault-tolerant capability, high in recognition rate, good in classification result and the like.

The invention provides an augmented reality unmarked tracking registration method and device based on an edge model, and relates to real-time performance and robustness of tracking registration in theaugmented reality unmarked field. At present, the defects that the surface of a three-dimensional object lacks enough textural features, pose jitter or disturbance is caused by few feature points, and the algorithm search space is too large are overcome. Aiming at these problems, according to the method, a linear parallel multi-mode LINE-MOD template matching method keeping rotation and scale unchanged is adopted to quickly identify a target object, and a reference view close to a current view angle is obtained to complete estimation of the pose of a camera. And the pose is used as an initialization pose based on an edge model tracking registration method, so that the initialization pose is prevented from being given manually. Therefore, the method can achieve the quick tracking and registration, and is good in real-time performance and robustness.

The invention provides a power line carrier (PLC) communication method and device. The method comprises the following steps: attenuating signal detection step: sending pilot frequency signals and receiving the pilot frequency signals after the pilot signals are transmitted by a power grid, and carrying out the attenuating signal detection on the received pilot frequency signals to acquire frequency bands with strong attenuation; interference signal detection step: collecting signals and carrying out interference signal detection so as to acquire frequency bands with strong interference; unionset taking step: taking a union set for the frequency bands with the strong attenuation and the frequency bands with the strong interference, wherein the frequency bands in the union set are frequency bands which are required to avoid; and communication step: modulating signals on frequency bands out of the frequency band union set, and carrying out normal communication. According to the invention, the interferences of signal attenuation frequency bands and interference signal frequency bands are avoided, and the instantaneity, the validity, the safety and the reliability in the process of data transmission are realized. The method and device provided by the invention are beneficial to the transmission of high-speed data, words, images and other multimedia service signals on transmission channels in the power grid.

The embodiment of the invention provides a method and a device for monitoring the power of an extra-high voltage connecting line. The method for monitoring the power of the extra-high voltage connecting line comprises the following steps of: acquiring a power error delta P of the extra-high voltage connecting line and an extra-high voltage area control error (ACE) in real time; in a two-dimensional plane, acquiring current coordinates by taking the power error delta P of the extra-high voltage connecting line as an abscissa axis and taking the extra-high voltage area control error (ACE) as anordinate axis, and acquiring an advancing curve of the power error delta P of the extra-high voltage connecting line and the extra-high voltage area control error (ACE) by taking time scales as a curve advancing source; and when the advancing curve enters a correspondingly preset extra-high voltage running state area, a correspondingly preset monitoring strategy of the extra-high running state area is operated. The embodiment of the invention can realize visualization real-time monitoring and pre-warning of extra-high voltage power control.

The invention discloses a vehicle supervision platform data access forwarding method and system, and the method comprises the steps that a vehicle-mounted terminal builds tcp connection with an accesssystem, the vehicle-mounted terminal sends a login request to the access system, the access system analyzes the login request, and the login request comprises a vehicle unique identification code vin; the access system verifies the vehicle-mounted terminal by verifying the legality and correctness of the unique identification code vin of the vehicle; if verification is unsuccessful, a tcp link isclosed; if verification succeeds, the vehicle-mounted terminal sends various acquired data to the access system for analysis; the analyzed data is original data and is not subjected to data offset, the access system analyzes the data and pushes the analyzed data to a kafka message queue, consumes the data from the kafka message queue, performs precision offset on the data, persists the data to adatabase, queries the database and displays the queried data through a page; and a forwarding system consumes the data in the kafka message queue, sends the data in an http form and forwards the datato other platforms.

A vehicle detection method comprises two parts: training of cascade multiple classifier and using the classifiers to carry out vehicle detection. The method is characterized by: using a LBP operator representation pixel level characteristic and a LBP histogram representation pixel block characteristic, and using a LBP spatial histogram with spatial information to characterize spatial structure information of the vehicle; using the LBP histogram characteristic to train a linear SVM classifier, using a boost method to train a second layer HIKSVM classifier, and detecting the vehicle according to a final classifier result. The LBP spatial histogram characteristic is used to characterize the vehicle information so that a vehicle outline can be rapidly characterized. Certain tolerance is possessed to a variety of an intra-class characteristic and strong distinguish is possessed to the variety of the outer class. The linear SVM classifier is used so that purposes of rapid training and rapid detection can be achieved. And real-time performance can be achieved too. The boost training method is used to train the second layer HIKSVM classifier so that a detection rate of the vehicle can be increased.

The invention discloses a rock slice image alignment method combining single polarized light and orthogonal polarized light images. The method comprises the following steps: S1, preprocessing a single polarized light image and an orthogonal polarized light image of a rock slice; s2, extracting feature points in the preprocessed image, and performing rough matching on the extracted feature points; s3, based on the rough matching result of the feature points, obtaining a homography matrix for pixel mapping by utilizing a bidirectional cross matching method, and taking the homography matrix as an accurate matching result; and S4, mapping all pixels of the orthogonal polarized image into the single polarized image based on an accurate matching result to achieve rock slice image alignment. According to the orthogonal polarized image and the single polarized image which are aligned through the method, a serious distortion phenomenon does not occur in a subsequent fusion process, and a relatively good image alignment effect is achieved; according to the method, the calculation time of image alignment can be reduced, the real-time performance is achieved, the influence of noise on feature extraction can be inhibited, and the method has stability when feature points are extracted.

The invention provides an image capturing device and image communication method which are applicable to video conference. The video conference is built between a local terminal and a remote terminal through network. The image capturing device comprises a processor, a video compressor, an encryption element and a data transmission interface, wherein the processor is used to process an original video signal to generate a first video signal and a second video signal; the video compressor is used to receive and compress the first video signal; the encryption element is used to encrypt the compressed first video signal; the data transmission interface is used to transmit the compressed and encrypted first video signal and the second video signal to the local terminal; and the first video signal is transmitted to the remote terminal through the local terminal and network and is fed back through the remote terminal, and the second video signal is displayed on the local terminal.