66results about How to "Meet the needs of real-time processing" patented technology

The invention discloses a moving target detecting and tracking method combining a frame difference method and a Gaussian mixture model. The method includes the following steps: acquiring a video image, and carrying out difference calculation on a current frame of input video image and a previous frame of video image to get an inter-frame difference image; using an initialized Gaussian mixture model algorithm to process the video image acquired to get a foreground region image and a background region image of the input video image; fusing the inter-frame difference image and the foreground region image to get a fused video image; binarizing and morphologically processing the fused video image to get a moving target image in the current input video image; and matching the moving target image extracted from the current input video image with a moving target image extracted from the previous frame of video image, and getting a moving target tracking result according to the matching result. By combining the advantages of two target detection algorithms, accurate and real-time detection of a moving region is realized.

The invention relates to an active-shape-model-algorithm-based method for analyzing a face expression. The method comprises: a face expression database is stored or selected and parts of or all face expressions in the face expression database are selected as training images; on the basis of the active shape model algorithm, feature point localization is carried out on the training images, wherein the feature points are ones localized based on the eyebrows, eyes, noses, and mouths of the training images and form contour data of the eyebrows, eyes, noses, and mouths; data training is carried out to obtain numerical constraint conditions of all expressions; and according to the numerical constraint conditions of all expressions, a mathematical model of the face expressions is established, and then face identification is carried out based on the mathematical model.

The invention discloses a video monitoring system and a moving target detecting and tracking method thereof. The method is applied to the video monitoring system on the condition that a camera is in a moving state. The method comprises the following steps: a moving target detection step, in which combination of a Harris corner and an HOG descriptor is employed as an image feature as well as processing including image feature extraction, image projection conversion, and image subtraction operation and the like is carried out to obtain a plurality of moving targets; a moving target model establishment step, in which moving target models are established for all the moving targets as well as a combination mode of a Harris corner, an extreme point and an HOG descriptor is employed to express features of the moving targets; and a target tracking step, in which target tracking is carried out according to the moving target models. According to the invention, corner feature matching is employed to reduce the calculated amount and precision of moving target detection is ensured; a corner and an extreme point are employed as target features, so that target identification has high robustness and tracking sustainability is strong; and a target motion estimation method is employed to reduce the calculated amount.

Provided is a method for realizing convolution operation based on a NOR Flash storage structure. The NOR Flash storage structure comprises a plurality of arrays of NOR Flash units. The convolution operation method comprises the following steps: the elements of the convolution kernel matrix are stored in the NOR Flash cell array; the elements of the input matrix are converted to voltages applied to the grid end of the NOR Flash unit; a driving voltage is applied at the source end of the NOR Flash units and the current value of each column is collected through the drain end of the NOR Flash units, thus convolution operation results are obtained.

The invention relates to a traffic monitoring video real-time defogging method based on moving targets. According to the characteristics of traffic monitoring videos, the contents of video frames are divided into foreground moving targets and background parts, and different methods are respectively adopted for carrying out defogging processing on the foreground and the background, wherein the foreground moving targets are mainly processed. Meanwhile, the relevance between the adjacent frames of the videos is utilized for adopting non-estimation or less-estimation methods on the background, and the processing speed is greatly accelerated. For 720*576 standard-definition videos, the processing speed being 25 frames / second can be reached, the real-time processing requirements can be met, and the problem that the defogging method in the prior art cannot meet the traffic monitoring video real-time processing requirement due to high processing complexity is solved. The defogging processing method provided by the invention has the advantages that higher pertinence is realized, the real color of the foreground moving targets can be more effectively remained, the color distortion can be reduced, and good foundation is laid for the further intelligent processing of traffic videos.

The invention provides a method and a system for monitoring a vehicle. The method comprises the steps of dividing each road into at least one road section according to the road information of each road and setting a unique road section identification and a corresponding road role for each road section, forming the running trajectory of the vehicle according to the running data of the vehicle, determining the road section identification of the road section on which the vehicle runs at present and the road rule corresponding to the road section identification according to the running data of the vehicle, and determining whether the vehicle gets out of the line according to the running data and the running trajectory of the vehicle and the road rule determined. The method is capable of performing whole-process dynamic real-time monitoring on the motor vehicles.

The invention provides a moving target detecting and tracing method and a system so as to solve the problem in the existing target detecting and tracing method that a plurality of false detection results easily appears when the scene is complex. When detecting a moving target, the method combines the background difference and the nuclear density estimating difference together and then conducts fusion processing to the image obtained by the two detection methods so as to eliminate the detection error of the two foreground images and finally pick up the moving target. Afterwards, the detected moving target is traced. The method and the system provided by the invention can better detect the variation of images under a complex scene with variational environment, thereby improving the accuracy of the detection. Furthermore, the method can handle the tracing problem of a plurality of moving targets simultaneously.

The invention provides a furnace mouth image preprocessing method. The method includes acquiring the furnace mouth flame image and removing the invalid regions on the periphery of the furnace mouth flame image to obtain an initial ROI; detecting the initial ROI and judging whether the invalid region with the longitudinal penetration exists or not; if the invalid region with the longitudinal penetration exists, segmenting the initial ROI into a plurality of sub-graphs; and removing the invalid sub-graphs in the sub-graphs to accurately extract the ROI. For the problem of image noise caused by the shielding of a furnace mouth area by a fireproofing door and a smoke cover under the complex light condition of a converter working site, the various irrelevant content and interferent objects canbe removed, and an effective ROI can be obtained, the recognition accuracy of the state of the converter can be obviously improved, the method has the advantages of being easy to realize and small incalculation amount, the requirements of real-time performance processing can be met, and the effectiveness of input images in the depth learning model can be guaranteed.

The invention discloses a number plate image preprocessing method, which comprises the following steps of: (1) obtaining the upper boundary and the lower boundary of a number plate character region; (2) determining a Yin and Yang boundary line of the number plate character region, wherein the Yin and Yang boundary line is a boundary line formed through brightness difference of a number plate; (3)calculating whether the number plate character region requires the brightness compensation or not; and (4) carrying out brightness compensation on the plate number character region. The number plate image preprocessing method has the advantages that the plate number identification rate of the Yin and Yang number plate can be obviously improved, in addition, the realization is easy, the calculation quantity is small, and the real-time processing requirement can be met.

The invention discloses a forming method of a concrete aggregate detection model. The method comprises the following steps: building a closed acquisition environment without sun illumination; in the closed acquisition environment, under the irradiation of a light source, acquiring an original image of the concrete aggregate by utilizing an image acquisition module; acquiring original images of different types of concrete aggregates, performing quality scoring on the original images to divide the original images into a plurality of grades, and taking the graded original images of the concrete aggregates as a deep learning training set; carrying out image preprocessing on the original images of the concrete aggregates in the deep learning training set; and training the deep learning trainingset subjected to image preprocessing by adopting a deep learning training model to obtain a concrete aggregate detection model. The invention further discloses a concrete aggregate quality detectionmethod. The method is high in detection precision and high in detection efficiency, can meet the requirement of real-time processing, and can also meet the requirement of stable detection in actual production or offline monitoring.

The invention discloses a method for achieving image defogging on an FPGA. The method includes the following steps of inputting a colored foggy image to the FPGA, obtaining pixel data of a channel R, a channel G and a channel B of the colored foggy image, storing the pixel data of the channel R, the pixel data of the channel G and the pixel data of the channel B into three read-only memorizers of the same size, conducting minimum value filtering on the pixel data of the three channels, taking the minimum value, namely, a dark channel, in the pixel data, conducting parallel calculation on the dark channel and the data of the channel R, the channel G and the channel B through a threshold value comparison method to obtain the overall-situation atmosphere light of the three channels, obtaining the transmitivity according to the overall-situation atmosphere light, the simplified dark channel and an original image, and conducting recovery to obtain a fog-free image according to the transmitivity, the atmosphere light and the original image. The method solves the problem that in the prior art, due to the fact that the foggy image processing speed is low, the real-time processing requirement can not be met.

A key frame extraction method of motion capture data comprises the following steps: (1) feature representation; motion components including rotation angles of a trunk and limbs of a human body and a relative distance between joints of the body are used to represent main features of a human motion posture; (2) critical point screening; an extreme point and a dynamic and static conversion point of a motion component curve serve as a critical point, and a critical point reduction algorithm is used to select a key critical point; (3) key degree curve structure; selected key critical points are used to divide a motion component curve, so as to form a motion component key degree curve; (4) weight learning; and (5) key frame extraction; weighting and fitting are used for a motion component key degree curve of a target motion, and the key frame extraction is carried out based on a fitting curve. The invention does not need to manually set the hard-to-determine threshold, can learn different key frame extraction styles from the sample motion, and can extract the key frame according to several limb motions, so that the need for real-time processing of the motion capture data can be met.

The invention discloses an image restoration method based on contour matching. The image restoration method based on the contour matching comprises the steps that a determined area to be restored is marked by marks, a block to be restored is defined in the imaged to be restored, a corresponding optimal matching block in a reference image is defined through the contour matching, pre-restoration is carried out on pixels to be restored in the block to be restored through the optimal matching block, the block to be restored is moved, the contour matching is carried out every time the block to be restored is moved to obtain optimal matching blocks, the pre-restoration is carried out on all the pixels to be restored in sequence, lastly, the pre-restoration is carried out on any pixel, weighted processing is carried out on pixel values of corresponding pixels in all the optimal matching blocks, and final restoration is carried out on the pixels by using the weighted processing results. According to the image restoration method based on the contour matching, matching restoration is carried out on the image to be restored through the contour matching, and the restoration is completed with the pixels as units step by step. Furthermore, the restoration comprises the pre-restoration and the final restoration, restoration precision is high through two-time restoration, needed manual intervention is less, and the image restoration method based on the contour matching is suitable for self-adaptation restoration.

The invention discloses an expression driving method and a device, electronic equipment and a storage medium, and relates to the field of artificial intelligence, and the method can comprise the steps: carrying out the face key point detection of a driven person in a first image, and obtaining a first face key point sequence; for each second image obtained in sequence, the following processing iscarried out: performing face key point detection on a driving person in each second image to obtain a second face key point sequence; analyzing the second face key point sequence if the no-expressionframe image is not determined currently, and taking the second face key point sequence as a no-expression key point sequence if the second image is determined to be the no-expression frame image basedon an analysis result; and otherwise, obtaining a deviation value between the second face key point sequence and the expression-free key point sequence, and performing expression driving rendering onthe driven person based on the deviation value and the first face key point sequence. By applying the method of the invention, the flexibility, interactivity, accuracy and the like can be improved.

The invention discloses an after-drilling inversion method for while-drilling electromagnetic wave well-log data of a high deviated well / a horizontal well. The inversion method comprises the followingsteps of s1, obtaining adjacent pilot well log data, and extracting a reservoir horizontal resistivity sequence; s2, based on while-drilling imaging data of the high deviated well / the horizontal well, determining stratigraphic interface position and tilt angle information; s3, performing layer-by-layer processing on divided stratums in the step s2, and building a simplified one-dimensional three-layer inversion model to reduce inversion parameters; s4, according to a change range of the reservoir horizontal resistivity sequence extracted in the step s1, applying geological constraints to theinversion parameters; s5, performing multi-initial value adaptive initialization on the inversion parameters; s6, performing iterative updating on a stratigraphic model by utilizing an adaptive multiplier regularization Gauss-Newton algorithm; and s7, optimizing and outputting an inversion result. The method improves the speed and precision of after-drilling inversion of while-drilling electromagnetic wave well-log of the high deviated well / the horizontal well.

The invention discloses a wide-angle lens calibration and image correction method. The method comprises the following steps: preparing a chessboard lattice image of 4*4 as a calibration template, acquiring a spherical image through a wide-angle lens shooting calibration template, performing RGB conversion YUV operation on the spherical image to obtain the grayscale image, processing the grayscaleimage to obtain an image containing nine border line angular points; the spherical radius is calculated by using the angular point information through an elliptic formula as a correction coefficient and the spherical image obtained in the lens use process is corrected by using the correction coefficient. The method is simple and convenient to calculate, can directly solve the calibration parameters and solve the corrected image by the embedded device, and meets the requirement of real-time processing.

The invention provides an image processing method based on multiple threads, computer equipment and a storage medium. The method comprises the steps that a main thread creates multiple decoding sub-threads; the main thread sends a decoding instruction to the plurality of decoding sub-threads, so that the plurality of decoding sub-threads decode each frame of image in each decoding task at the sametime according to the decoding instruction to obtain a target decoded image set corresponding to each decoding task; the main thread controls a graphic processor to perform image recognition on eachframe of image in the target decoded image set to obtain a detection result set of the target decoded image set; the main thread performs data analysis on all detection result sets obtained in the current time period to obtain an image analysis result of each decoding task in the current time period; therefore, the problems of low processing speed and low processing quality of a video image processing method in the prior art are solved, and the requirement of a user for real-time processing of the video image is met.

The invention discloses a signal lamp image processing method, device and equipment and a storage medium, and belongs to the technical field of image processing. The method comprises the following steps: acquiring a long exposure image and a short exposure image obtained by shooting a signal lamp; carrying out halo suppression processing on the long exposure image, carrying out format conversion,and obtaining a first image, wherein the first image is a red, green and blue RGB image; carrying out format conversion on the short exposure image, carrying out color adjustment processing, and obtaining a second image, wherein the second image is an RGB image; and carrying out fusion processing on the first image and the second image, and carrying out format conversion to obtain a signal lamp image after correction processing. By fusing the first image and the second image, the corrected signal lamp image is obtained, the contour and color of the signal lamp can be accurately recovered, thesignal lamp of any color is corrected, the correction effect of the signal lamp image is improved, the universality is improved, the calculated amount is small, and the processing efficiency is high.

The invention provides a pointer representation number identification method and device, and electronic equipment. The method comprises the steps of obtaining a to-be-identified pointer table image; extracting a color region containing a pointer and scale lines in the HSV color space, setting a result as mask, and segmenting a pointer region on the mask; connected components of the pointer area are extracted; determining a rotation rectangle corresponding to the connected component of the pointer area; determining a scale line area based on the rotation rectangle and the mask; and identifyinga reading corresponding to the pointer table image based on the rotating rectangle and the scale line area. According to the method, color prior information is utilized to extract color areas of a pointer and scale lines in an HSV color space, a dial pointer area and a dial scale line area are determined, and finally readings corresponding to a pointer table image are identified through the pointer area and the scale line area. The method is high in recognition precision, strong in scene adaptation robustness and simple in algorithm, can effectively shorten the operation time of the algorithm,and can meet the requirements of real-time processing.

The present invention relates to a data processing system is provided which comprises at least one first processing unit (CPU), at least one second processing unit (PU), at least one memory module (MEM), and an interconnect. The memory module (MEM) serves to store data from said at least one first and second processing unit (CPU, PU). The interconnecting means couples the memory module (MEM) to the first and second processing units (CPU, PU). In addition, an arbitration unit (AU.) is provided for performing the arbitration to the memory module (MEM) of the first and second processing units (CPU, PU). The arbitration is performed on a time window basis. A first access time during which the second processing unit (PU) has accessed the memory module and a second access time which is still required by the second processing unit (PU) to complete its processing are monitored during a predefined time window by the arbitration unit (AU). The second access time is compared to the remaining access time in the time window and if the remaining access time is larger than the second access time, the arbitration unit (AU) allows said at least one first processing unit (CPU) to access the memory module in said time window. Otherwise, the arbitration unit (AU) restricts the access of the at least one first processing units (CPU) and allows the at least one second processing unit (PU) to access the memory module (MEM).

The embodiment of the invention provides a video processing method and device, and a handheld camera, and the method comprises the steps: firstly determining a target image recognition algorithm froma plurality of candidate image recognition algorithms according to the priority calculation information corresponding to the candidate image recognition algorithms; and carrying out image processing on the to-be-processed image frame in the video by using a target image recognition algorithm. Therefore, according to the embodiment of the invention, not only can corresponding image processing be carried out on the image frames in the video by using multiple image recognition algorithms to meet diversified video processing or description requirements, but also the image processing time is shortened due to the fact that the to-be-processed image frames are only subjected to image processing by using the target image recognition algorithm, and the image processing efficiency is improved. And the requirement of real-time processing in the video shooting process can also be met.

The invention discloses an image processing method, and the method comprises the steps: carrying out the processing of a reference image, and obtaining a target resolution reference image; obtaining a target resolution ratio motion residual according to the motion residual of a to-be-processed image block in the to-be-processed image relative to a reference image block in the reference image; and obtaining a target resolution image block of the to-be-processed image block according to a target resolution reference image block and a target resolution motion residual in the target resolution reference image. By the adoption of the method, the requirement for resolution processing of the image in the video is met.

The invention discloses a real-time track obstacle detection method based on a three-dimensional point cloud, and the method comprises the steps: carrying out the processing of three-dimensional point cloud sequence data collected by a laser radar, firstly carrying out the coordinate transformation of the point cloud, converting the coordinates in a European coordinate system into the coordinates in a spherical coordinate system, putting each point in the point cloud into a certain voxel of a cone by using a cone voxelization down-sampling method so as to reduce the calculation amount of subsequent steps; inputting the downsampled points into a local feature coding module, searching local point clouds by using K-nearest neighbor (KNN), aggregating geometric features of the local point clouds, and connecting the centroid, neighbor point coordinates, relative coordinates and Gaussian density features of the local point clouds into a vector; connecting all local point cloud information into a matrix through traversal, and obtaining high-dimensional local feature information of each local point cloud through MLP and maximum pooling; and finally, utilizing multi-scale three-dimensional sparse convolution to realize track real-time identification of a single-frame image through a plurality of down-sampling and up-sampling modules.

The invention provides a real-time identification method and a real-time identification system for a pesticide spraying sector angle of an unmanned aerial vehicle. The method comprises the steps of acquiring a real-time image which comprises a pesticide applying nozzle and a pesticide applying spraying sector; extracting a green component in the real-time image, and using the green component as an objective image; acquiring a central pixel of the objective image; according to the central pixel, acquiring a pixel a and a pixel b which are separated from the central pixel by m pixel distances in a vertical line which passes through the central pixel; acquiring a first horizontal line which passes through the pixel a and a second horizontal line that passes through the pixel b; acquiring a right upper side boundary point, a right lower side boundary point, a left upper boundary point and a left lower side boundary point; and acquiring the sector angle of the pesticide spraying sector according to the right upper boundary point, the right lower side boundary point, the left upper boundary point and the left lower side boundary point. The real-time identification method and the real-time identification system can realize quick identification for the pesticide spraying sector angle and furthermore satisfy a requirement for real-time performance.

The invention relates to a run length-based one-scan connected component labeling method and hardware structure. The method comprises the steps of (1) simultaneously scanning two adjacent rows of an input binary image, and recording equivalent run length between the two rows; (2) updating an equivalent run length table after the scanning is completed; (3) detecting a completed connected component,and writing the run length of the completed connected domain into an external memory; and (4) if the current row is the last row, writing all the run lengths in the row and a run length buffer into the external memory, otherwise, repeatedly executing all the steps.

The invention belongs to the technical field of underwater acoustic physics, and particularly relates to a multi-beam seabed sedimentary layer type estimation method based on transfer learning, and the method comprises the steps: taking a part of data in multi-beam echo intensity data on a target sea area survey line as transfer data, and inputting the transfer data into a pre-trained deep convolutional neural network model, performing fine tuning on partial network node parameters of the deep convolutional neural network model to obtain a corrected neural network model; and inputting multi-beam echo intensity data received in real time on a target sea area survey line into the corrected neural network model, outputting a multi-beam seabed sedimentary layer type, and realizing estimation of the multi-beam seabed sedimentary layer type.

The invention discloses a method for combining reading data in an image distortion correcting algorithm, wherein the image distortion correcting algorithm is a bilinear interpolation algorithm. A current point is subjected to interpolation operation according to a pixel value of a window where the current point is, and each row of pixels is provided with a plurality of continuous current points. The method is characterized in that some row of a plurality of current pixel data and the next operation row of a plurality of current points needed by the interpolation operation of the current point are read in a combining way. Based on the method, a real-time modifying function can be realized by using an FPGA (Field Programmable Gata Array) of an SPANTAN6 and a hardware platform of DDR2SDRAM with the monolithic data rate of 400MHz*16bit, under the condition that the polynomial resolution of barrel distortion distortion model with the HD (High-Definition) video 1080p30 data bit 16bit YUV 4:2:2 meet formulas: K3=0.7 and K5=0.5, so that the hardware platform is simplified.