143 results about "Video reconstruction" patented technology

A system for capturing live-action three-dimensional video is disclosed. The system includes pairs of stereo cameras and a LIDAR for generating stereo images and three-dimensional LIDAR data from which three-dimensional data may be derived. A depth-from-stereo algorithm may be used to generate the three-dimensional camera data for the three-dimensional space from the stereo images and may be combined with the three-dimensional LIDAR data taking precedence over the three-dimensional camera data to thereby generate three-dimensional data corresponding to the three-dimensional space.

A camera for acquiring a sequence of frames of a scene as a video includes a sensor with an array of sensor pixels. Individual sensor pixels are modulated by corresponding modulation functions while acquiring each frame of the video. The modulation can be performed by a transmissive or reflective masked arranged in an optical path between the scene and the senor. The frames can be reconstructed to have a frame rate and spatial resolution substantially higher than a natural frame rate and a spatial resolution of the camera.

The invention discloses a quasi-three dimensional reconstruction method for acquiring two-dimensional videos of static scenes, which belongs to the field of computer vision three-dimensional video reconstruction. The method comprises the following steps: step A, extracting double-viewpoint image pairs from each frame of a two-dimensional video; step B, respectively and polarly correcting each double-viewpoint image; step C, adopting a binocular stereo matching method based on overall optimization to respectively solve overall optimum disparity maps of all the polarly corrected double-viewpoint images; step D, reversely correcting the overall optimum disparity maps so as to obtain the corresponding disparity maps of all the frames in the three-dimensional video; step E, splicing the disparity maps obtained in the step D according to a corresponding video frame sequence to form a disparity map sequence, and optimizing the disparity map sequence; and step F, combining all the extracted video frames and the corresponding disparity maps, adopting a depth image based rendering (DIBR) method to recover virtual viewpoint images, and splicing the virtual viewpoint images into a virtual viewpoint video. The method is low in computational complexity, simple and practicable.

The invention relates to an Adaptive frame structure-based AVS coding method and relates to a method for processing video signals. According to the method, the scene switch is detected by reading a video coded image frame by frame, and the relevance and the complexity of the adjacent frames are analyzed at the same time; coding types of the images to be coded is judged, and a coding order of the images is adjusted to determine the selection of the reference frames. By considering the characteristics of video sequences, the method ensures the compression efficiency of the video images, can greatly improve the subjective quality and the stability of the AVS video reconstruction images, and eliminate a mosaic appearance caused by the AVS coding in the process of scene switch or violent motion. The method can be used in the video sequences in one scene or a plurality of scenes.

The invention discloses a fluid-solid interaction simulation method based on video reconstruction and an SPH model. The fluid-solid interaction simulation method includes steps: 1), adopting a bright-dark recovery shape method to quickly reconstruct fluid surface geometric information in video, and acquiring a surface height field of each frame of input image; 2), combining the height fields with a shallow water equation, and calculating to acquire a speed field on the surface of fluid in a form of a minimizing energy equation; 3), using the surface geometric information as a boundary constraint condition to be discretized into a whole three-dimension body to acquire volume data; 4), guiding the volume data reconstructed in the video into an SPH simulation scene to serve as an initial condition of the simulation scene, and interacting with other virtual environment objects in the scene. By the fluid-solid interaction simulation method, high-accuracy data reconstruction can be realized, fluid surface details can be retained, bidirectional interaction simulation is performed on the basis of reconstructed data and a physical simulation model, fluid animation effect closer to real condition is acquired, algorithm complexity is low, and the method has high creativity compared with related algorithms.

The invention discloses a video loop filtering method based on a deep convolutional network. The method comprises the following steps: step 1, making a training data set for loop filtering network training; step 2, constructing a network model for video filtering; step 3, taking the training data set obtained in the step 1 as a training set of the network, respectively training two models for intra-frame prediction frame filtering and inter-frame prediction frame filtering, and forming a video filtering network model by the two models; training the video filtering network model by taking a minimized loss function as an optimization target; and step 4, integrating the video filtering network model trained in the step 3 into video encoding software to complete the whole video encoding process and obtain a reconstructed frame after passing through the video filtering network. Compared with a traditional filtering method, the method has the advantages that the image quality of the video reconstruction frame is improved, the accuracy of inter-frame prediction is improved, the coding efficiency is improved, the filtered image frame has higher reconstruction quality, and the video codingefficiency is greatly improved.

The invention belongs to the technical field of cross-media generation, and particularly relates to a video description generation system based on a graph convolution network. The video description generation system comprises a video feature extraction network, a graph convolution network, a visual attention network and a sentence description generation network. The video feature extraction network performs sampling processing on videos to obtain video features and outputs the video features to the graph convolution network. The graph convolution network recreates the video features accordingto semantic relations and inputs the video features into sentence descriptions to generate a recurrent neural network; and the sentence description generation network generates sentences according tofeatures of video reconstruction. The features of a frame-level sequence and a target-level sequence in the videos are reconstructed by adopting graph convolution, and the time sequence information and the semantic information in the videos are fully utilized when description statements are generated, so that the generation is more accurate. The invention is of great significance to video analysisand multi-modal information research, can improve the understanding ability of a model to video visual information, and has a wide application value.

The invention discloses a nonuniform sparse sampling video super resolution method, which belongs to the technical field of video super resolution. The method comprises key technologies such as shot detection, key frame extraction, original image fuzzy processing and down sampling, nonuniform sampling vision model building, sparse dictionary construction, and video reconstruction. A nonuniform sparse sampling method based on foveated vision is adopted, differential sparse sampling is carried out on a video sequence, the number of atoms is greatly reduced, a dual dictionary of high and low resolutions (high and low resolution reference copy) is generated in real time or calculated in a prior mode according to the resolution of a mobile terminal device screen as carried meta data, a hardware-supported nonlinear Mipmap interpolation method is used for simulating and generating a Foveation image in real time, a result similar to that in a Gauss pyramid method can be acquired, and the computation overhead is lower.

The invention discloses a variable-length input super-resolution video reconstruction method based on deep learning. The method comprises the following steps: constructing a training sample with a random length, and obtaining a training set; establishing a super-resolution video reconstruction network model, wherein the super-resolution video reconstruction network model comprises a feature extractor, a gradual alignment fusion module, a depth residual error module and a superposition module which are connected in sequence; training the super-resolution video reconstruction network model by adopting the training set to obtain a trained super-resolution video reconstruction network; and sequentially inputting to-be-processed videos into the trained super-resolution video reconstruction network for video reconstruction to obtain corresponding super-resolution reconstructed videos. According to the method, a gradual alignment fusion mechanism is adopted, alignment and fusion can be carried out frame by frame, and alignment operation only acts on two adjacent frames of images, so that the model can process a longer time sequence relationship, more adjacent video frames are used, that is to say, more scene information is contained in input, and the reconstruction effect can be effectively improved.

The invention discloses a positioning method of repeated fragments based on video reconstruction. The positioning method includes the steps of removing mass redundancy among video frames in a query video sequence and a quote video sequence, and extracting representative video key frames for small video fragments which have close time and consistent contents to achieve a downsizing process of data; describing contents of the video key frames by utilizing pseudo text documents; converting visual similarity calculation among video key frames into similarity measurement among pseudo text documents, measuring the similarity between query video key frames and quote video key frames, and returning a series of similar quote key frames for each query key frame; comprehensively using time consistency restraints of repeated video fragments and time inconsistency among non-repeated fragments to reconstruct quote video fragments and determine boundary information of the repeated video fragments. The positioning method is applicable to mining of data media and copyright protection.

The invention discloses a fluid simulation method based on video reconstruction and eulerian model coupling. The fluid simulation method comprises the steps of 1) reconstructing a three-dimensional density field of each frame of a fluid according to an input video; 2) solving an N-S equation by adopting an eulerian method, and updating the speed field and density field of the fluid; 3) taking the reconstructed density field of adjacent two frames as prior information and taking the result of the eulerian method as correction to reconstruct a three-dimensional speed field of the fluid; and 4) guiding the eulerian fluid simulation by using the reconstructed density field and speed field to generate a new animation effect. By adoption of the fluid simulation method, the density field and the speed field of the fluid can be reconstructed with relatively high precision; and by tightly coupling the reconstructed data and a fluid geometrical model, a fluid animation effect which is more closer to an actual condition can be obtained, and controllable fluid details can be added.

The invention discloses a time-space video compressed sensing method based on a convolutional network, and mainly aims at solving the problems that in the prior art, the video compression time-space balance and the video reconstruction real-time performance are poor. The scheme of the method comprises the steps that a training data set is prepared; a network structure of a time-space video compressed sensing method is designed; training and testing files are written according to the designed network structure; a network of the time-space video compressed sensing method is trained; and the network of the time-space video compressed sensing method is tested. The network of the time-space video compressed sensing method adopts an observation technology of simultaneous time-space compression are conducted simultaneously and a reconstruction technology of using 'time-space blocks' to enhance the time-space correlation, not only can real-time video reconstruction be achieved, but also the reconstruction result has the high time-space balance, the reconstruction quality is high and stable, and the network can be used for compressed transmission of a video and follow-up video reconstruction.

The invention provides an adaptive partition compression and perception-based video compression method. The method comprises two steps, namely the step of adaptively partitioning video images, and classifying and assigning sampling rates for various image blocks. During the step of adaptively partitioning video images, a gray difference value between the adjacent pixels of a reference frame image is adopted as a basis for block size segmentation, and a partitioning threshold value T is set. The gray average difference value between the adjacent pixels of a current region block is compared with the threshold value, and the adaptive blocks of the video image are partitioned based on the quad-tree algorithm. In this way, flat regions are effectively separated from detail regions and edge regions. On the basis of the adaptive partitioning operation, an inter-frame difference value for the DCT coefficients of video pixels is adopted as a basis for partitioning, and various image blocks diversified in size are divided into three types, namely quickly changing blocks, transition blocks and slowly varying blocks. Meanwhile, appropriate sampling rates are assigned to different types of image blocks. The method is good in video reconstruction quality and short in reconstruction time. Under the same condition, the video reconstruction quality and the reconstruction time of the above method are better than those of the video uniform partitioning, compression and perception processing method.

The invention provides a staring super-resolution imaging device and method, and aims at solving the problem of low resolution of an existing imaging device. The device comprises an imaging device lens group (1), a detector (2), a detector driving platform (3), a driver (4), a storage device (5), an image processing circuit (6) and an image display circuit (7), wherein the detector (2) is arranged in a focal plane on a light path of the imaging device lens group (1); a drive part of the driver (4) drives the detector driving platform (3) to generate movement with magnitude and direction being random values. The imaging method comprises the following realization steps: a change number instruction is set; the driver (4) drives the detector (2) to do corresponding number of random jitter; a light signal is imaged on the detector (2) through the imaging device lens group (1) to obtain a low-resolution image with a frame number corresponding to a change number; the low-resolution image is rebuilt by the utilization of a variational Bayesian algorithm. The staring super-resolution imaging device and method are suitable for video reconstruction and satellite photography.

The invention discloses a DCT-based 3D-HEVC fast intra-frame prediction decision-making method. The method comprises the following steps: firstly, calculating a DCT matrix of a current prediction block by utilizing a DCT formula; then, judging whether the left upper corner coefficient of a current coefficient block is provided with an edge and further judging whether the right lower corner coefficient thereof is provided with an edge; and finally, judging whether DMMs need to be added into an intra-frame prediction mode candidate list through judging whether the edges are provided. According to the DCT-based 3D-HEVC fast intra-frame prediction decision-making method, a depth map is introduced into the 3D-HEVC to achieve better view synthesis; aiming at the depth map intra-frame predicationcoding, a 3D video coding extension development joint cooperative team proposes four new kinds of intra-frame predication modes DMMs for the depth map. The DCT has the characteristic of energy aggregation, so that whether a coding block is provided with an edge can be obviously distinguished in the 3D-HEVC depth map coding process. The DCT-based 3D-HEVC fast intra-frame prediction decision-makingmethod has the advantages of low calculation complexity, short coding time and good video reconstruction effect.

The invention discloses a joint regularization based video super-resolution reconstruction method, which comprises the steps of building a low-resolution observation model; building a universal video reconstruction cost function through a regularization based least mean square method; constructing a compensation based TV (CTV) regularization item, and assigning an adaptive weighting coefficient to a regional space so as to reduce adverse effects brought about by a registration error; constructing a multi-frame nonlocal low-rank (MNLR) regularization item; building a joint regularization based video super-resolution reconstruction cost function; solving the cost function by using a Split-Bregman iteration method so as to reconstruct high-resolution video. Video frames reconstructed according to the video super-resolution reconstruction method have abundant edge information, and hardly have any sawtooth effects. It is observed from the video reconstruction frames that the method disclosed by the invention is also excellent in ability of noise suppression, and the method has a very high reference value in an objective evaluation parameter. Therefore, disclosed by the invention is an effective video super-resolution reconstruction method.

The invention provides a dual-band time domain compressed sensing high-speed imaging method and device. Aiming at a high-speed moving object, the characteristic of overturning of a digital micromirrordevice is utilized, and the light energy in two directions is respectively used for imaging at visible light and infrared bands; in combination with a time domain compressed sensing imaging method, the detail information at double wavebands of the high-speed moving target is obtained at the same time, and a video reconstruction algorithm based on compressed sensing is used to reconstruct the shotdouble-waveband low-speed video into a double-waveband target high-speed video. According to the invention, the method can obtain the high-speed moving target video with more detail information whilethe data conversion bandwidth of the imaging camera is not increased, thereby solving the problem of the strict requirements of high-speed imaging for high sensitivity and ultrahigh data bandwidth ofthe camera. Meanwhile, through the characteristics of the optical modulator of the digital micromirror device, the visible light and infrared high-speed imaging light paths can be obtained at the same time under the condition that no additional device is added, the light energy is fully utilized, and the utilization rate of the energy is improved.

The embodiment of the invention provides a video transmission method and device, electronic equipment and a readable storage medium. Video source data in a video cache queue is encoded; correspondingrandom access frames in different video cache queues are enabled to be acquired image frames at the same moment; adding a parameter set containing the code rate, the spatial resolution, the time resolution and the random access frame time interval of a data code stream obtained by encoding of each video encoder in a plurality of video encoders at the same time; therefore, the sending end can sendthe data in the case of bandwidth fluctuation; the method comprises the following steps: receiving packet loss information of current video reconstruction data obtained by decoding sent by a decoder and network time delay data of the current video reconstruction data reaching the decoder; and switching data code streams in real time in combination with the parameter set, so that a decoder decodesthe data code streams to obtain video reconstruction data, and the video reconstruction data displays clearer and smoother pictures.

The invention provides a video semantic communication method based on hierarchical knowledge expression, and mainly aims to solve the problems of incomplete semantic extraction, insufficient semantic representation capability and redundant semantic description in the prior art. According to the implementation scheme, the method comprises the following steps: constructing a hierarchical knowledge base consisting of a multi-level signal sensing network, a semantic abstract network, a semantic reconstruction network and a video reconstruction network; collecting a to-be-transmitted video signal; extracting structured semantic features of the video signals based on a signal sensing network and a semantic abstract network in the hierarchical knowledge base, and transmitting the structured semantic features through an ultra-narrowband channel; and reconstructing a video signal by using a semantic reconstruction network and a signal reconstruction network in the hierarchical knowledge base according to the structured semantic features. According to the method, semantic features of different scales are mined, and the semantics are structurally represented by using the structured data, so that the integrity of semantic extraction is improved, the semantic representation capability and the communication bandwidth utilization rate are improved, and the method can be applied to online conferences, human-computer interaction and intelligent Internet of Things.

Disclosed is a method of detecting a moving object in video produced in a plurality of CCTV cameras. In the present invention, a video reconstruction is applied to the CCTV storage video in order to significantly reduce the amount of video data, and then an image processing is performed on the reconstructed storage video in order to detect a moving object and further to extract metadata thereof. According to the present invention, staff members of the integrated control center may efficiently search any possible critical objects in the CCTV storage video.

The invention discloses a reliable video transmission method based on network coding to mainly solve the problem of lower video reconstruction quality of a receiving end caused by packet loss or error codes in a general video transmission method. The reliable video transmission method based on network coding comprises the following steps of: adding corresponding redundant information to a layered code stream used for performing the scalable video coding for each picture group of original video information, and packing the layered code stream to complete code rate distribution; performing the random linear network coding on data packets subjected to code rate distribution by using a designed global coding matrix with a nonstringent lower triangular structure; and transmitting coded information to the receiving end through a transmission channel, and decoding a network coding code stream at the receiving end by using a Gaussian elimination method, and restoring the original video information. The reliable video transmission method based on network coding has the advantages of improving the problem of lower video reconstruction quality of the receiving end caused by packet loss and time delay in the transmission process, and realizing the unequal error protection of an H. 264/SVC (scalable video coding) code stream and the high-reliability video multicast.