65 results about "Background reconstruction" patented technology

The invention discloses a characteristic matching and MeanShift algorithm-based target tracking method. The method comprises the following steps: inputting an image sequence, carrying out background reconstruction on the image sequence, obtaining a target area of an initial moment and modelling by adopting a MeanShift algorithm; carrying out SIFT characteristic extraction on a target area model of the initial moment and taking the SIFT characteristic point of the target area model of the initial moment as an initial characteristic point of a characteristic library; calculating the initial position, size parameter and rotation parameters of the current frame of target through SIFT characteristic matching; accurately positioning the target by adopting the MeanShift algorithm; calculating a shielding factor of the target, judging the shielding degree of the target and determining the tracking mode of the target; and ending the target tracking after all the images in the image sequence are tracked. According to the method, the MeanShift algorithm and the SIFT characteristic matching algorithm are combined and the advantages of the two algorithms are exploited so that the stable tracking the target is realized.

The invention relates to a mobile phone screen defect detection method based on regular texture background reconstruction. The method comprises the following steps of firstly, carrying out Fourier transform on an image and calculating an amplitude spectrum of an image frequency spectrum; carrying out Hoff straight line fitting on a binary amplitude spectrum so as to acquire a filtering template; using the template to carry out filtering on a real portion and an imaginary portion of the image; then carrying out Fourier inverse transformation and normalization so as to acquire a background reconstruction graph without defects; and finally using an original graph to subtract the background reconstruction graph, and carrying out adaptive binarization so as to acquire a binary image which only contains the defects. By using the method, a screen defect under a regular texture background can be effectively positioned.

The invention belongs to the field of machine vision and defect detection, and particularly relates to a mobile phone screen defect segmentation method, device and equipment based on a converged network. The method comprises the following steps of acquiring mobile phone screen images including a defect image and a defect-free image, training a pre-established defect detection network by using thedefect image and using a transfer learning method, and obtaining a defect candidate box corresponding to the defect image, training a pre-established image reconstruction network by using the defect-free image, and recovering a background reconstruction image, performing difference operation on the defect image and the background reconstruction image, and obtaining a defect segmentation image by adopting a threshold segmentation mode, utilizing the position coordinates of the corresponding defect candidate boxes on the defect segmentation image to extract the corresponding defect parts of thedefect segmentation image under the position coordinates, and obtaining a final defect segmentation result. According to the method, the defect detection network and the image reconstruction network are combined, so that not only can a small defect target be detected, but also the defect image can be accurately segmented.

The invention discloses a method for detecting and shooting a vehicle based on a composite virtual coil. The method comprises the following steps of: (1) carrying out background reconstruction on an image sequence of a traffic video to obtain a background image in a monitored region of a vidicon; (2) generating the composite virtual coil; (3) extracting the region of the vehicle; (4) updating a vehicle driving information list; and (5) storing three images for recording the illegal process of the vehicle according to the position relationship between the region of the vehicle represented by each item in the vehicle driving information list and the virtual coil so that an illegal vehicle is shot. According to the method for detecting and shooting the vehicle based on the composite virtual coil disclosed by the invention, the problems of inflexible time selection and low shooting accuracy of the traditional virtual coil method in the process of shooting can be solved; the disadvantages of incapability of detecting and shooting vehicles pressing lane lines and illegally steering in the traditional virtual coil method can be overcome; and, compared with the traditional virtual coil method, the method disclosed by the invention has the advantages of flexible shooting time selection and high shooting accuracy and is capable of detecting and shooting vehicles driving not according to lanes.

The invention provides a method for analyzing the synchronism of double-people diving, which mainly comprises: extracting movement targets on the basis of dynamic background reconstruction in movement videos of double-people diving, representing and extracting the synchronism characteristics on the basis of synchronism grading factors in diving rules, and adopting a bias character statistic learning method to construct synchronism evaluation functions to carry out synchronism evaluation. The method can automatically extract athlete external outlines in the movement videos of double-people diving. The method provides a method for effectively representing the synchronism characteristics according to the diving rules. The invention also introduces the sequencing idea which is normally used in bias character statistic learning into the problem of constructing the synchronism evaluation functions of the movement videos of double-people diving, and the absolute score problem is converted into the relative ordering problem. Finally, the synchronism of double-people diving can be automatically estimated through calculating the synchronism evaluation function value of the movement videos of double-people diving. The method can estimate the synchronism of double-people diving accurately, effectively and automatically.

The invention discloses a method for structurally reconstructing a background for intelligent video monitoring. The method comprises the following steps that: an image is acquired by an image frame and is updated by an updating module; a picture movement optical flow estimation module detects full-picture movement optical flow of the image; a picture segmenting module further segments a dynamic inter-frame picture; an image mean value module uses a static region as a mask template; a partial background reconstructing module fills a pixel mean value of the static region to a corresponding position of a background picture; a background reconstruction state detecting module calculates reconstruction percentage of the background picture and judges whether reconstruction is finished, if not, the first step is carried out again so as to acquire a new image frame, if so, the background is output by an output background module. By adopting the method disclosed by the invention, when frame image data are acquired, images acquired after a certain time interval are very high in reconstruction accuracy and are free from reconstruction error, and a static background picture of a supervised scene is reconstructed rapidly, efficiently and clearly.

The invention discloses a moving object detection method with background reconstruction based on neighborhood correlation, which comprises the following steps of: inputting an image sequence, and sequencing data; dividing gray scale stable region classes; calculating the occurrence frequency of each gray scale stable region class; dividing background unstable areas, and determining a candidate background for pixel points; determining a background of pixel points; and detecting a moving object. The invention has the advantages that the amount of calculation is less; a model is not required for the background and objects in a scene, and condition assumption is not required for the background; the background can be reconstructed from a scene image with a moving prospect, and thus, a mixing phenomenon can be avoided effectively; a satisfied result can be obtained in a large range of parameter variation; a background can be reconstructed accurately for an area of which the background does not occur in the maximum frequency; and the robustness is good. The invention has wide application potential in the field of real-time systems, such as machine vision, video monitor, military science, urban traffic monitoring, resident routine safety monitoring, and the like.

The invention provides a video synopsis generation method capable of solving the multi-target collision and occlusion problem. The method comprises a background reconstruction module, a target detection module, a target tracking module, a target trajectory post-processing module, a target trajectory generation module and a video synopsis generation module. The method is an object-based dynamic video synopsis generation method; the method carries out intelligent analysis processing on surveillance videos, carries out detection and tracking on targets in a source video and extracts trajectory information of target positions and sizes and the like, and generates a concentrated video comprising all of the targets through trajectory combination and optimization; and the method can reduce video space redundancy and time redundancy.

The invention discloses a low-rank video background reconstructing method, mainly solving the problem that a video background image cannot be clearly and reliably reconstructed when a video image sequence is subjected to background reconstruction in the prior art. The low-rank video background reconstructing method is realized by the following steps: firstly, carrying out low-rank decomposition on an input video to be processed X, so as to obtain an initial background estimation image GL; secondly, detecting a target area in the video X to be processed; setting all pixel values of the target area to be 0, and setting a first image in the video X to be processed as a reference image; using pixel values of the other images to fill the target area in the reference image, so as to obtain a background estimation image G; and finally, using a grey degree value of the initial background estimation image GL to replace the target area which is not entirely filled, so as to obtain a background image. With the adoption of the low-rank video background reconstructing method provided by the invention, when the video sequence is subjected to the background reconstruction, the clear and reliable background image can be obtained. The low-rank video background reconstructing method is applicable to the background reconstruction of the video sequence with various fixed backgrounds.

The invention discloses a Gaussian mixture-based moving target detection method. The method comprises the following steps of 1) building a Gaussian mixture model; 2) obtaining a changed region through difference, and performing measurement and analysis judgment on an area of the changed region; 3) analyzing and judging whether light sudden change occurs in the changed region or not; 4) adopting different moving target detection methods according to a judgment result of the step 3); and 5) detecting out a moving target. According to the method, the moving target is detected by selectively adopting a background difference method or an inter-frame difference method through measuring and judging the area of the changed region and judging whether the light sudden change occurs in the changed region or not by adopting a background reconstruction mechanism updated in a cyclic period; and the method well solves the problem that a conventional background subtraction method is relatively sensitive to the illumination change, also solves the problem that a relatively long-time virtual shadow is easily generated during conversion between a foreground and a background in a scene, and has relatively good robustness.

The invention discloses an automatic leucorrhea trichomonad detection method based on moving target recognition, and belongs to the field of image processing. The method comprises the steps that sample images are continuously acquired and then processed to acquire foregrounds and backgrounds of the images, trichomonad areas are recognized at the positions where the foregrounds and the backgrounds are changed by conducting foreground and background analyzing on the continuous images, and then trichomonad is recognized according to the morphological characteristics of the trichomonad. Therefore, the method has the advantages that a traditional Kalman background reconstruction method is improved, noise brought by illumination changing, lens shifting and focal length changing can be effectively inhibited, and the background and foreground areas can be accurately partitioned.

The invention discloses a calligraphy background reconstruction method based on image super resolution. The method is characterized by, to begin with, selecting a background region having less noise in a color image of a calligraphy background and carrying out background reconstruction on the selected background region; and meanwhile, segmenting background, text and seal information in calligraphy work, and carrying out pixel-level fusing on the background, text and seal information to obtain a final reconstructed image. The method well solves the problem of loss of artistic information of a conventional denoising method, and improves completeness of the artistic information of the calligraphy work; and besides, the method effectively solves the problem that a conventional method cannot deal with the noise of ink spreading and natural weathering and the like, so that a high-quality calligraphy background image can be obtained, and ornamental value of the artistic information of the calligraphy work is improved.

The invention provides a depth-map gaining method of plane videos. The depth-map gaining method comprises the following steps of: detecting and extracting a foreground moving object in an original plane video, calculating a blocking relation of the foreground moving object and obtaining a mask image of the foreground moving object; carrying out background reconstruction on the original plane video according to the mask image of the foreground moving object so as to obtain a background video sequence with removal of the foreground moving object; solving a background depth-map sequence for the background video sequence; solving an initial depth-map sequence for the original plane video; and obtaining a depth-map sequence of the original plane video according to initial depth information of the foreground moving object in the initial depth-map sequence and geometric information of the foreground moving object in the background depth-map sequence. According to the method provided by the embodiment of the invention, the depth map of each frame in the plane video can be accurately gained; the gained depth map is clear in edge, explicit in depth hierarchy, good in smooth performance and high in time-domain stability.

The invention belongs to the field of computer vision, and provides a cascaded convolutional neural network fused with space-time clues, which is used for realizing video foreground and background separation. Therefore, the technical scheme adopted by the invention is as follows that in a video foreground and background separation method based on a cascaded convolutional neural network, two encoder-decoder type sub-networks are used to carry out video foreground and background separation; the two sub-networks are respectively an FD network for foreground detection and a BR network for background reconstruction, the FD network is used for generating a binarized foreground mask, and the BR network reconstructs a background image by using output and input video frames of the FD network; in order to introduce a space clue, three continuous video frames are taken as input; in order to improve the network applicability, the optical flow graph corresponding to the original video frame is usedas a space clue and is input into the FD network at the same time. The method is mainly applied to occasions of video foreground and background separation.

The invention relates to an image-based shopping recognition and ideographic expression method. On the basis of the logic design architecture, face recognition matching is realized by means of multi-frame comparison, so that the face recognition accuracy rate is increased; for users after successful matching, background reconstruction and foreground object extraction are carried out to obtain an object image and the obtained object image is trained and identified, so that the object feature is recognized well. During the recognition process, the background interference can be removed, so thatthe object held by the user is identified; and the hinting action of the user is identified to complete object purchasing. The entire process is realized automatically, efficiently and fast.

The invention discloses a sparse coding background modeling method. The method adopts a discriminative atom model method for carrying out background modeling on images, and comprises the following steps that 1, collected images are divided into a plurality of image blocks, and the image blocks are subjected to sparse coding; 2, on the basis of the sparse coding model of the images, discriminative atoms are found out from atoms in a sparse coding dictionary by using a frequency-inverse file frequency (tf-idf) statistics analysis method; 3, the selected discriminative atoms are used for completing the image background rebuilding. The technical scheme is adopted, so the image background modeling method provided by the invention has the advantages that the average information quantity and frequency-inverse file frequency (tf-idf ) technology is utilized for carrying out statistics analysis on the atoms in the sparse coding dictionary on the basis of the image sparse coding model, the atoms carrying the discriminative information, i.e. the discriminative atoms are found out, and the discriminative atoms are used for rebuilding the image background information.

The invention discloses a screen defect detection method, device and system, computer equipment and a medium. A specific embodiment of the method comprises the steps of inputting a to-be-detected screen image into an enhanced learning network to generate an intermediate image; inputting a target training set formed by the defect-free screen image and the intermediate image into a discriminator ofthe generative adversarial network to generate a discrimination result, feeding back the discrimination result to the reinforcement learning network as a return value of the reinforcement learning network until the discrimination result meets a preset convergence condition, thereby obtaining a background reconstruction image; and differentiating the to-be-detected screen image and the background reconstruction image to obtain a defect image. According to the embodiment, background reconstruction of the to-be-detected screen image can be performed based on the mutually constrained reinforcementlearning network and the generative adversarial network so that the defect image capable of clearly presenting the defect position and the defect quantization level can be obtained, and visual and accurate defect detection can be realized.

The invention discloses a moving target detecting method based on background reconstruction, comprising the following steps: inputting an image sequence and computing the gray level difference of the adjacent frames of the pixels; classifying the stable gray level intervals; computing the mean gray level of each class of stable gray level intervals; merging the similar classes of stable gray level intervals; selecting the background gray levels of the pixels; and detecting the moving target. The method saves storage space, is small in computed amount, good in robustness and wide in application range, dispenses with establishing models of the backgrounds and targets in the scenes and can directly reconstruct the backgrounds from the scene images containing moving foreground and effectively avoid mixing. The method has extensive application prospect in such real-time system fields as machine vision, video monitoring, military science, urban traffic monitoring, daily resident safety monitoring, etc.

The invention discloses a hyperspectral image target detection method based on sample mining and background reconstruction, and mainly solves the problem of low target detection precision in the prior art. The method comprises the steps of performing coarse detection on an input hyperspectral image, and obtaining a training sample based on a coarse detection result; respectively constructing a generative adversarial network, a reverse auto-encoder network and an auto-encoder network, and respectively training the generative adversarial network, the reverse auto-encoder network and the auto-encoder network by using training samples; calculating a reconstruction error and a preliminary detection result of the autoencoder network for reconstructing the input hyperspectral image; obtaining an optimized hyperspectral image and a feature map according to the preliminary detection result, and further realizing second-stage sample mining, network training and target detection to obtain a second-stage detection result; and fusing the preliminary detection result and the second-stage detection result to obtain a final detection result. The method can make full use of background spectrum information, effectively inhibits background interference, improves the target detection precision, and can be used for environmental protection, mineral exploration, crop yield estimation and disaster prevention and relief.

A method for detecting a moving target of an ink-jet printing fabric based on a mixed-state Gauss MRF model comprises the steps: (1) inputting an observation video and setting iteration implementation parameters; (2) calculating an iterative optimization solution of a grid point and generating the mixed-state Gauss MRF state value of the grid point according to grid point state judging policy; (3) if the grid point is a moving target point, marking the value of a moving target detection diagram at the grid point and keeping the background reconstruction value of the grid point constant; or else, updating the background and setting the new background reconstruction value of the grid point; (4) generating the state points of all grid points by iteration and generating a mixed-state Gauss MRF state set by utilizing the state points; (5) calculating an overall energy value by utilizing the mixed-state Gauss MRF state set; (6) calculating overall energy change value; if the change value is greater than an iterative error threshold value, continuously performing ICM (Iterated Conditional Modes) iterative optimization; or else, finishing the convergence process of the iterative optimization; and (7) outputting the moving target detection diagram and a background reconstruction diagram. The dynamic updating of the background in the moving target detection process can be realized, the representational capacity of the complex texture background can be improved effectively, the detection precision under the noise environment is improved and the method is applicable to the detection treatment for the moving target of the ink-jet printing fabric.

The invention discloses a compressed sensing based video background reconstruction and emergent mass incident early warning platform which comprises a video input unit, a compressed sensing based background detection module, a change detection module, a pedestrian detection and number counting module, a report generation module and an early warning determination module. The early warning platform uses video data collected by video collectors as the basis, a video analysis method based on compressed sensing is utilized, background information in a warning area is reconstructed in an L1 norm minimizing optimization algorithm, mobile objects in the video can be detected, influence of inhuman objects as vehicles and animals in a scene is eliminated in a pedestrian identification algorithm, the number of pedestrians in the video images is counter in real time, and if a preset maximal allowed number is exceeded, the alarm is raised in real time, police and related staff timely arrive at the area to disperse the crowd and maintain the order, and accidents as disturbance can be avoided.

The invention discloses an infrared and visible light image fusion method based on saliency fusion. Three extraction methods including salient target extraction based on infrared image segmentation and Bessel background reconstruction based on a quadtree, sparse salient target extraction based on structure low-rank coding and salient target multi-scale detail extraction based on Laplace are adopted, an infrared salient target is determined, and the infrared image and the visible light image are fused, so that the visual saliency characteristic of the target is obtained, and meanwhile, the detail information of the scene is reserved. According to the image fusion method provided by the invention, the defects of the infrared image can be overcome, and the advantages of visible light and infrared light can be exerted, so that the fused image has the advantages of infrared light and visible light at the same time, and the detection and reconnaissance capability of the system is improved.

The invention discloses a self-adapting background subtracting method based on Gaussian mixture background reconstruction. The method comprises steps of collecting video frames, extracting an initial background frame, and initializing a background model; establishing Poisson distribution of a noise model by using R (red), G (green) and B (blue) component differences of the current frame and the background frame, counting a histogram of the Poisson distribution, and calculating relative variances for the obtained histogram; and ranking the obtained relative variances, finding a maximum value to serve as a segmentation threshold of R, G and B components of the current frame, conducting binaryzation and obtaining a foreground frame. The method is adapted to dynamic background perturbation and light change effect, moving objects in a video can be detected in real time, and the method is good in robustness.

The invention discloses an intelligent monitoring video processing method which comprises the following steps: (1) video preprocessing: converting a video into a static image, then filtering noisy points in the video image, and meanwhile, carrying out offset adjustment on the data of the whole image, so that the pixel gray scale distribution is uniform; (2) foreground extraction: carrying out background reconstruction by utilizing a static empty scene image, then adding a system space identifier according to a preset region, carrying out background subtraction on a new image, and extracting apixel region with relatively large difference as an activity foreground; (3) behavior tracking: carrying out convolution processing on the extracted activity foreground, and taking a foreground contour; extracting feature points, and recording a movement track; and (4) result analysis: analyzing a final result, and correspondingly storing various types of data according to various service requirements so as to meet various service function requirements. The method disclosed by the invention can assist safety personnel to deal with crisis more quickly and effectively, and the phenomena of falsealarm and missing alarm are reduced to the maximum extent.

The invention discloses a texture surface defect detection method and system based on abnormal synthesis and decomposition, and belongs to the field of image processing. According to the method, a segmentation-guided defect generation network is constructed, a large number of defect samples similar to real defects can be generated by using a small number of real defect training samples, and meanwhile, an anomaly synthesis method based on Gaussian sampling is provided, and anomaly negative samples can be randomly synthesized by using defect-free positive samples, so that the problem of small quantity of defect samples in industry can be solved; the defect detection precision is further improved; according to the method and system, the abnormal negative sample is decomposed into the texture background image and the abnormal mask image by adopting the abnormal decomposition network, so that defects can be effectively prevented from being reconstructed into the texture background, the texture background reconstruction precision is improved, the defect area can be accurately segmented, and the residual image and the abnormal segmentation image are fused; therefore, the defect detection rate is improved, and the defect over-detection rate is reduced.

The invention discloses a lane line detection method, device and system based on background reconstruction used for detecting lane lines in a road image quickly and efficiently. The lane line detection method comprises the steps that lane line features in a to-be-detected road image are acquired based on background reconstruction; image processing is performed on the lane line features, and a processing result is obtained; and Blob analysis is performed on the processing result to confirm lane line image blocks on the to-be-detected road image. According to the method, background reconstruction is performed on the to-be-detected road image, the lane line features are quickly acquired, an algorithm process of road lane line detection is simpler and more efficient, and therefore quick detection of the lane lines is realized.

The invention discloses a video OSD extraction and coverage area reconstruction method. According to the method, video analysis, front and back frame time calculation, position calculation, interpolation and other methods are adopted to perform information extraction and correction, and video reconstruction and restoration are performed on an original OSD shielded area after information extraction. According to the method, OSD information extraction of the video monitoring image is realized through view analysis; time calculation is carried out through comparison of front and back frame identification numbers; oCR identification accuracy threshold alarm is set; calibration data collected by means of front and back frame verification, identification accuracy early warning and human assistance is used as a sample set for OCR model training. And the OSD information is used as a view to be sent to a background view in a structured mode for data mining of view big data, and background reconstruction of an OSD area is achieved through interpolation of a front and back frame interval average method. According to the method, the identification accuracy and the credibility of OSD information extraction are improved.

The invention discloses a static video analysis method and system. The method comprises the following steps: acquiring video data; obtaining a linear dynamic regular term of the video data background;obtaining a structured sparse regular term of the foreground of the video data; obtaining a sparse regular term of the noise; according to the linear dynamic regular term, the structured sparse regular term of the foreground and the sparse regular term of the noise, constructing a decomposition model combining a dynamic background and structure sparsity; and performing optimization solution on the decomposition model to obtain a foreground and background separation result of the video data. The static video analysis method provided by the invention has very good background reconstruction capability and foreground detection performance under most challenging conditions, has the advantages of detection universality, high accuracy and strong robustness, has self-adaptive capability to a dataacquisition environment, and can effectively eliminate the influence of adverse factors such as noise and illumination.

The invention discloses a target detection method based on background reconstruction. The method comprises a step of determining a value range of N to be a range from ft to 2f according to the a time t of detection system initialization and a frame frequency f of an image acquisition device in the detection system, a step of obtaining continuous N frames of video images from a k-(N-1) frame to a current frame k and obtaining a reconstruction background image through taking a pixel average value, a step of using a background difference method to cut images for the reconstruction background image and the current frame k images, and carrying out morphological filtering to obtain a background difference image B (k), a step of calculating a difference image D(k) of images of a kth frame and a (k-r) frame with r as a frame difference, wherein r is larger than or equal to 5, and a step of determining a target position through comparing the background difference image B (k) and the difference image D(k). By using the method, an ideal background model can be obtained, and thus the accuracy of a detection result is improved.