187 results about "Frame difference" patented technology

The invention relates to the technical field of image communication, in particular to a computer network communication data processing method and system based on artificial intelligence, and the method comprises the following steps: carrying out the continuous frame difference calculation of the data of real-time video frames, collecting the pixel-level change data between two frames, and obtaining a difference data set; according to the method, continuous frame difference calculation is carried out on real-time video frames, pixel-level change data between two frames is captured, storage and transmission occupation of useless data is reduced, and data processing pressure is relieved. And the compression of the difference data further optimizes the size of the data packet, so that the transmission efficiency is improved, and the integrity of video key content is guaranteed at the same time. The compressed difference data packets are analyzed, key visual elements are extracted, and limited transmission resources are distributed to important information parts in a centralized manner by focusing on visual key points such as human faces and actions, so that the pertinence of the transmission process is enhanced.

The invention discloses a complexity and semantic perception video analysis method and system based on edge cloud collaboration. According to the method, video preprocessing is optimized, continuous similar video frames are efficiently processed, video semantic features and system workload changes are considered, and an effective scheduling strategy is achieved. According to the method, a difficulty estimator and an inter-frame difference detector are constructed, the processing difficulty and redundancy degree of a task are predicted according to semantic features of a video task, a state is generated in combination with current system measurement (queue length and bandwidth), and an optimal decision is made through a decision engine based on a PPO algorithm, and the optimal task resolution and the final processing node are selected. And after reasoning, the processing reward of the current task is calculated, and the DRL model is updated according to the reward to continuously adapt to the changes of video contents and system loads, so that the overall performance is improved to the greatest extent, and the reasoning quality, the processing speed and the resource utilization rate are balanced.

The invention relates to the technical field of video source switching, and discloses an intelligent video source switching system and method for a director switching station. The method comprises the following steps: acquiring an input video stream of a current video source through a director switching station, and performing N-gram feature vector calculation and inter-frame difference analysis to obtain video frame segmentation data; performing scene analysis on the video frame segmentation data to obtain a scene feature signature; performing spatial-temporal feature extraction and fusion on each video frame in the input video stream according to the scene feature signature to obtain a scene spatial-temporal fusion feature set; inputting the scene space-time fusion feature set into an LSTM-based lightweight time sequence convolutional network model to carry out video switching behavior prediction so as to obtain video switching behavior prediction data; and inputting the video switching behavior prediction data into a double-depth Q learning model based on a Markov decision process to carry out video source switching decision to obtain a video source switching decision result, thereby improving the intelligent video source switching accuracy of the director switching station.

The invention discloses a traffic video question and answer method based on inter-frame difference, and the method comprises the steps: innovatively constructing an inter-frame difference extraction module on the basis of an image pre-training model based on the local abrupt change characteristic of a traffic video: effectively capturing the abrupt change characteristic of a moving target through the pixel-level difference calculation of adjacent frames; and designing a gating feature fusion mechanism, dynamically aligning spatial distribution of differential features and visual features on the basis of fusing problem features, and realizing collaborative characterization of static scenes and dynamic changes. In order to enhance cross-modal feature alignment, a hierarchical comparison learning strategy is provided: differential-visual feature comparison is implemented at a bottom layer to enhance local dynamic perception, and text-visual comparison is performed at a high layer to establish global semantic association. Experimental results show that the classification performance is improved on the basis of the SUTD-TrafficQA reference data set.

The invention provides a multi-identity internal aggregation deep counterfeit video detection method, which comprises the following steps of: extracting shallow texture difference information and shallow frequency domain information through a modified Officient-B0 network, designing a random channel single-head attention mechanism to carry out feature fusion, mining inter-frame difference information by using space attention and time attention, and carrying out multi-identity internal aggregation on the inter-frame difference information. The problem of deep counterfeit video detection in a multi-identity complex scene is solved.

The invention discloses a vehicle monitoring method based on frame difference and deep learning fusion, and relates to the technical field of vehicle monitoring, cameras and environment sensors are deployed in a monitoring area, and videos and multi-source data are acquired by means of vehicle-road cooperation; a self-adaptive frame difference method is used, morphology and optical flow estimation are matched, a threshold value is determined according to the environment, and vehicle features are extracted; constructing a deep convolutional neural network with an attention mechanism, and training a model by using various data in combination with migration and reinforcement learning; fusing the two types of features based on a graph attention network to form high-quality fusion features; a space-time diagram convolutional network is combined with an LSTM to track a vehicle and predict a trajectory, a behavior pattern library is constructed to judge abnormity, and classification analysis is performed in combination with an SVM and a knowledge graph. According to the invention, the frame difference and deep learning are fused, the monitoring accuracy is improved, and the vehicle can be accurately identified and detected; the real-time performance is enhanced, the data is quickly processed, and the environmental influence is reduced; traffic management is assisted, and a safe and efficient traffic environment is created.

The invention relates to the technical field of physical examination scoring, in particular to a solid ball examination method and system based on AI, and the method comprises the following steps: based on a solid ball throwing video of an athlete, employing an inter-frame difference method, calculating the pixel change between continuous video frames, recognizing a motion area, separating a foreground from a background, and determining the positions and movement of the athlete and the solid ball; and generating motion area data. According to the method, the dynamic positions of the athlete and the solid ball can be accurately identified and tracked by applying the inter-frame difference method and the optical flow method, so that more accurate and continuous movement track information is provided, and key actions are identified and classified more accurately by using the time convolutional network and the time sequence clustering; accuracy and efficiency of action analysis are improved, and more objective and meticulous evaluation is realized on technical execution quality scoring and comprehensive technical evaluation through application of a support vector machine and a decision tree.

The invention relates to the technical field of termite monitoring data processing, in particular to an intelligent analysis method and system for monitoring data of a visual termite monitoring device.The method comprises the steps that the visual termite monitoring device is designed and used for monitoring termite activities; the method comprises the following steps: acquiring a video through a visual termite monitoring device, calculating a background difference value by adopting a frame difference method, dynamically updating a background template, and extracting a termite activity frame; in combination with a termite detection model, obtaining the coordinates and number of activity bounding boxes, calculating the moving direction and speed of termites, generating termite data and uploading the termite data to the cloud; termite data are obtained through the cloud, the movement index, the activity index, the frequency index and the activity degree are calculated, and meanwhile, the termite activity trend in the future is predicted through an activity prediction model in combination with environment data. Through real-time monitoring and activity prediction, the first alarm information and the second alarm information are respectively sent out, comprehensive analysis and accurate early warning of termite activities are realized, and thus the monitoring efficiency is improved.

The invention discloses a to-be-detected image screening method, device and equipment and a computer readable storage medium. The method comprises the following steps: determining an effective detection time period corresponding to each camera; acquiring video data acquired by each camera in the corresponding effective detection time period; for each path of video data, detecting whether a differential operation result of adjacent image frames in the video data is greater than a first threshold; and if the image frame is greater than the first threshold value, taking the image frame acquired later in the adjacent image frames as an image to be detected. According to the method and the device, the video data acquired by each camera in the corresponding effective detection time period is taken as a basis, and the frame difference algorithm is combined to screen the to-be-detected image so as to be used for a subsequent detection task, so that the data volume faced by the detection task is reduced, and the computing resources are saved.

The invention relates to the technical field of resolution improvement, in particular to a video definition improvement method and system.According to the video definition improvement method and system, through window accumulation of adjacent frame gray difference and coordinate difference, minimum displacement is selected, mapping is updated, boundary neighborhood weighted correction is carried out, cross-frame dislocation is reduced, and motion area connection is guaranteed; capturing a three-frame gray scale sequence to calculate a difference value, writing a replacement value according to a monotone and amplitude condition, inputting the replacement value and an alignment frame difference into Fourier transform, decomposing amplitude and phase distribution in a residual field in a frequency domain, and performing amplitude and phase consistency elimination on an abnormal block during frame-by-frame superposition to obtain an abnormal block; row difference and column difference absolute values of neighborhood blocks in the residual iteration texture matrix are accumulated into energy, the energy and a center difference are combined with a threshold value to output a mark, and a texture region and an edge region are distinguished; the brightness of the low-frequency position is updated according to the weight mean value, the maximum difference and the direction difference are combined for the high-frequency position, the edge gradient is extracted through the convolutional neural network, a correction value is generated, edge steps and fuzziness are suppressed, and high-frequency details are continuously kept.

The invention relates to a static and dynamic feature fusion video and text cross-modal retrieval method. The method comprises the following steps: constructing a noun, preposition and verb set for a text; converting nouns and prepositions into text static feature representation Tsf through a text encoder, and converting verbs into text dynamic feature representation Tdf; encoding a video through a video encoder to obtain video static feature representation Vsf; calculating and normalizing an inter-frame difference sequence of adjacent frames of the video, and encoding through a video encoder to obtain a video dynamic feature representation Vdf; calculating the static similarity between the Tsf and the Vsf, calculating the dynamic similarity between the Tdf and the Vdf, and carrying out weighted average on the Tsf and the Vsf to obtain a final similarity score; for a plurality of given texts and videos, k text-video pairs with the highest similarity scores are selected as retrieval results. According to the method and the system, the accuracy and the efficiency of video and text cross-modal retrieval can be improved, and videos related to text description can be effectively retrieved from a large amount of video data.

The invention discloses a time domain filtering method and device based on video inter-frame difference, equipment and a medium, and relates to the technical field of video coding. The method comprises the following steps: acquiring a to-be-filtered frame and a reference frame corresponding to the to-be-filtered frame, and dividing the to-be-filtered frame and the reference frame to obtain video blocks; determining an inter-frame difference corresponding to each video block in the to-be-filtered frame; the inter-frame difference is the difference between two video blocks at the same position in the frame to be filtered and the reference frame; and determining a filtering weight corresponding to the inter-frame difference according to the difference threshold, and performing time domain filtering on the corresponding video block in the to-be-filtered frame based on the filtering weight. Therefore, the filtering weight is determined by combining the inter-frame difference, and the time domain filtering intensity can be adjusted according to the inter-frame difference, so that the time domain filtering is performed on the to-be-filtered frame by adopting the proper filtering intensity, the time domain filtering capability is improved, and the phenomenon that the code stream is not reduced and then is increased is avoided.

The present invention belongs to the technical field of machine vision detection, and relates to a method for detecting sizing yarn hitting a reed based on machine vision. The present invention provides a method for detecting sizing yarn hitting a reed based on machine vision, which can detect and track the phenomenon of warp yarn hitting a reed during the sizing process, and is a method for detecting sizing yarn hitting a reed with high accuracy, strong robustness, and wide variety adaptability. The complete image of the warp yarn passing through the reed teeth is captured by an image acquisition system that is independently built and can adapt to different types of sizing machines. First, the initial image is cropped to obtain the detection area, and then the image is smoothed by Gaussian filtering; finally, by improving the inter-frame difference method, expanding the inter-frame difference, adding the Canny edge algorithm and mathematical morphology operation, the detection and tracking of the warp yarn hitting the reed are realized; at the same time, the proposed algorithm can be applied to detection and recognition problems in other fields.

The invention relates to the technical field of robot control, in particular to a robot decision-making method based on deep reinforcement learning, which comprises the following steps: acquiring a change rate data sequence, calculating a frame difference value, screening data frames greater than a threshold value, extracting a mean vector, generating a feature vector set, splicing a ratio of remaining time to distance, constructing a coding matrix and adjusting weight parameters. Screening task groups, mapping and sorting, generating a task batch set, screening adaptive tasks, and generating an execution sequence list. According to the method, adjacent frame difference screening is carried out on multiple change rate sequences in the environment perception data, the mutation section features are extracted, the strategy priority is dynamically adjusted by screening specific inner product results in the strategy evaluation process, the strategy adaptability is improved, the task combination with the minimum conflict and the highest priority is screened, and the strategy evaluation efficiency is improved. And task batches with higher matching degree are screened according to the calculation capability difference value, so that the autonomous response capability and execution reliability of the robot system in a complex task scene are enhanced.

Systems, apparatus, and methods for post-processing video e.g. frame denoising. Noise reduction techniques may be employed to improve the quality of digital video. Frames may be extracted from a video. Synthetic frames may be created using motion data between the extracted frames. Synthetic frames may be masked to exclude pixels from the composite frame. Thresholds used in masking may vary based on the temporal distance of the extracted frame used to create the synthetic frame and the extracted frame. Masking may be based on frame differences between extracted and synthetic frames (e.g., sub-pixel / luminance differences), areas of lower quality motion data (e.g., occlusions), or edge detection in the extracted frames. Synthetic and extracted frames may be composited generating frames having less noise. The composited frame may be based on averaging pixel values across the synthetic and extracted frames. Composited frames may be compiled and encoded into denoised video.

The invention discloses a process pipeline micro-vibration motion amplification detection method based on automatic tracking and deconvolution, and the method comprises the steps: carrying out the automatic frame selection of a vibration abnormal region through an inter-frame difference method, tracking displacement through combining with an LK optical flow method, and analyzing a frequency spectrum, so as to adaptively determine an amplification frequency band; performing vibration signal enhancement by selecting an Euler or phase video motion amplification technology according to the environmental complexity; and further converting the RGB image into an HSV color space, inhibiting motion blur and artifacts by adopting confocal deconvolution, and finally synthesizing a high-definition amplified video. The method has the advantages that automatic recognition of the vibration area and the frequency band is achieved, and the detection efficiency is improved; the image quality is remarkably improved through deconvolution post-processing, and noise interference is avoided; complex industrial environments are compatible, reliable technical support is provided for non-contact pipeline health monitoring, leakage and explosion accidents are effectively prevented, and the service life of facilities is prolonged.

The invention discloses an automobile long-distance speed recognition and tracking method based on target detection, and the method comprises the steps: collecting a vehicle driving video in real time through a fixed camera position and angle, and carrying out the recognition of a vehicle in each frame of image of the video through a pre-trained YOLOv12 target detection model; performing preliminary speed estimation on the automobile by adopting an inter-frame difference method, further constraining a speed change rate, and performing time sequence correction on the speed change rate by adopting Kalman filtering to obtain a corrected speed estimation value; calculating the displacement and the motion deflection angle of the adjacent frames in the transverse and longitudinal directions, and integrating the displacement and the motion deflection angle with the corrected speed estimation value to form a time sequence input feature; and carrying out modeling on a historical track through the trained time sequence deep learning model to realize dynamic speed correction. According to the method, the dependence on physical space calibration is obviously weakened, the perception capability of the model on the motion trend is improved, and the method is suitable for vehicle tracking and speed measurement tasks in long-distance and low-resolution scenes.

The invention relates to a monitoring video stream transmission method and device, equipment and a medium. The method comprises the following steps: acquiring performance parameter data, hardware resource utilization rate data and network condition data of video monitoring equipment and acquired video stream data; calculating a brightness difference in the video stream data by adopting a frame difference method; analyzing the spatial and temporal distribution characteristics of the brightness difference to obtain a motion significance index; determining the priority of video stream transmission according to the motion saliency index; based on the priority, combining the performance parameter data and the hardware resource utilization rate data, and adopting a first preset rule to allocate a video coding strategy for each video monitoring device; and generating a monitoring video stream transmission scheme by adopting a second preset rule according to the video coding strategy, the network condition data and the current hardware resource utilization rate data. According to the invention, the transmission priority and the coding strategy can be dynamically adjusted according to the video content, and the transmission efficiency and the quality stability of the monitoring video stream are improved.

The invention provides a slope rockfall monitoring and early warning method, device and equipment based on dynamic depth estimation and a medium. Relates to the technical field of civil engineering operation maintenance. The method comprises the following steps: S1, acquiring a video sequence image; s2, performing dynamic target detection on the video sequence image by adopting a frame difference method; s3, based on a pre-trained rockfall identification model, identifying rockfall in the target frame; s4, based on a depth-pro monocular image depth estimation model, performing depth estimation on the rockfall in the target frame to obtain a rockfall position distance; s5, performing feature analysis on the rockfall in combination with the depth data to obtain feature parameters; s6, performing tracking and trajectory identification on the rockfall; and S7, performing alarm reminding according to tracking and trajectory recognition results. According to the method, the extracted features are trained to identify the slope rockfall, the position and the size of the rockfall can be identified more accurately, a sharp measurement depth map is rapidly provided, and the dynamic change of the slope rockfall is captured in time.

The invention provides an early-stage enteromorpha identification method, medium and system, and belongs to the technical field of enteromorpha identification. According to the early-stage enteromorpha identification method and system, a dynamic feature fusion module based on inter-frame difference is used for extracting time sequence features of continuous frames, cross-modal information integration is achieved through a double-attention multi-modal feature fusion module, an improved YOLOV8 network model is constructed for target detection, and the identification accuracy of the early-stage enteromorpha is improved. Meanwhile, a liquid neural network model is combined with a multiband ratio method and a floating algae exponential function to carry out spectral feature recognition, enteromorpha components in mixed pixels are processed through a sub-pixel decomposition technology, and finally, multi-dimensional information such as the detection confidence coefficient, the spectrum matching degree, the enteromorpha abundance value and the target number is comprehensively analyzed and detected through a decision fusion algorithm. Accurate recognition and positioning of the early enteromorpha are achieved, and the technical problem that accurate recognition and distinguishing of the early enteromorpha are difficult to achieve at present is solved.

The invention discloses a jamming frame detection method and system for games, and relates to the technical field of computers, and the method comprises the following steps: fusing multi-modal features by using PyTorch, generating time sequence representation through TCN modeling based on the multi-modal features, mapping into jamming probability through a multilayer perceptron structure based on the time sequence representation, and optimizing the jamming probability by using a Focal Loss loss function. And a lagging detection result is generated. The frame difference threshold is dynamically optimized by adopting DQN reinforcement learning, so that the complexity of a game scene can be self-adapted, and the false alarm rate and the omission ratio are remarkably reduced; multi-modal features are fused through a multi-scale SE attention mechanism and PyTorch, and the feature expression ability and the model robustness are effectively improved; tCN time sequence modeling is used to accurately capture a lagging related time sequence mode, the detection precision under unbalanced data is optimized, and the accuracy and practicability of lagging detection in a high-dynamic game scene are enhanced.

The invention discloses a spray head synchronous calibration method based on computer vision. The method comprises the following steps: S1, generating a dense three-dimensional depth point cloud model containing a relative position relation between a spray head array and a substrate; s2, performing inter-frame difference calculation on the binocular difference images of the continuous frames; s3, outputting a current global six-degree-of-freedom pose estimation result of the nozzle array in real time through a joint residual function; s4, constructing a nozzle-level emission time offset calculation model based on the overall six-degree-of-freedom pose estimation result of the nozzle array; s5, generating a corresponding space compensation control instruction based on the three-axis translation deviation and the three-axis rotation deviation contained in the nozzle array global six-degree-of-freedom pose estimation result; s6, through the synergistic effect of the time compensation control in the step S4 and the space compensation control in the step S5, nozzle time synchronous calibration and array space pose synchronous calibration are achieved. According to the invention, global pose information of the nozzle array can be output in real time, and dynamic capture of continuous inter-frame tiny pose changes is realized.

The present invention relates to the field of big data technology, and discloses an urban rail transit emergency handling method and system based on a visual large model, including: collecting visual data, sound data, and equipment operation data at the platforms and carriages of urban rail transit, performing preprocessing and timestamp synchronization; performing dynamic frame difference processing on consecutive video frames of the visual data to generate a two-dimensional motion heat map, and performing spatio-temporal aggregation; constructing a three-dimensional digital model based on the engineering drawings of the platforms and carriages, and establishing a mapping relationship between the two-dimensional motion heat map and the three-dimensional digital model; performing abnormal behavior detection on the mapping results, and performing multi-modal fusion analysis in combination with the sound data and equipment operation data to determine the severity of the abnormal event; formulating an emergency response strategy according to the severity level of the abnormal event, sending control instructions to the rail transit equipment through an industrial protocol interface, and pushing the abnormal event information to the control center, inspection personnel, and passenger terminals.

The invention discloses an embedded firmware camera data modification method and device, a medium and equipment, and relates to the technical field of computer vision, and the method comprises the steps: collecting image data in real time, and carrying out the primary processing; a frame difference method and a background modeling algorithm are adopted to carry out region division and priority distribution on the image data after primary processing, and a region priority mapping table is generated; extracting feature points of a high-priority motion area in the optimized image data through an optical flow method, and calculating a motion vector between continuous frames; according to the calculated motion vector between the continuous frames, an event-driven frame processing mechanism is triggered, and the frame rate and the processing precision are automatically adjusted; through the motion vector extracted through the optical flow method, the motion speed of the object can be calculated in real time, the frame rate and the processing precision are dynamically adjusted according to the motion speed, it is ensured that enough details are captured in the rapid motion process, and processing resources are saved in the static process.

The invention relates to the technical field of video compression transmission, in particular to a high-reliability network video playing control system, which comprises a data acquisition module used for acquiring network bandwidths at all moments in all transmission time periods and the number of videos being transmitted in all the transmission time periods in the transmission process of network videos; the parameter correction module is used for acquiring the corrected quantization parameter, and comprises the following steps: extracting each connected domain of each frame of image in the network video; obtaining a matched connected domain, a change factor and a contour evaluation value corresponding to each connected domain; calculating inter-frame difference degree and element richness of each frame of image; calculating the network instability, the compression coefficient and the corrected quantization parameter of each transmission time period; and the compression and transmission module is used for carrying out real-time compression and transmission on the network video by combining the corrected quantization parameter with an HEVC algorithm. By dynamically adjusting the quantization parameters, the transmission reliability and the broadcast control fluency of the network video in complex scenes such as a weak network are improved.

The invention discloses a multi-algorithm fusion detection method and device for a dynamic target and a medium, and relates to the technical field of image processing, and the method comprises the steps: collecting continuous video frames of a moving target in a backlight imaging environment; adopting a three-frame difference method to extract dynamic target candidate areas in continuous video frames; on the basis of the structure-enhanced YOLOv8-P2 model, target detection result extraction is carried out on the candidate region under the condition that low-layer feature information is reserved; performing grid slicing on an original video frame in the continuous video frames at a preset overlapping rate, and inputting the original video frame into the RT-DETR model to perform target detection result extraction under background noise suppression; and two target detection results with different emphasis are fused based on spatial position alignment, and coordinates and confidence information of the dynamic target are output. According to the method, advantage complementation is formed through a double-model parallel architecture, and the adaptability of the system in harsh industrial environments such as backlight and dynamic impurities is remarkably improved on the premise that the real-time processing capacity is maintained.

The invention relates to a motion detection method and system based on a continuous frame difference method in combination with a CNN, and belongs to the technical field of digital image processing. The method comprises the following steps: acquiring a video frame sequence in real time through a camera, and performing differential operation on video frames of the continuous video frame sequence to extract a potential motion area; inputting the extracted potential motion area into a convolutional neural network for deep feature extraction and recognition; and analyzing the extracted depth features, judging whether the motion belongs to an abnormal motion behavior or not, and if the motion is detected to be the abnormal motion behavior, automatically triggering an alarm by the system and recording related data. According to the method, the CNN processing range is reduced, the processing speed is remarkably improved, the changed part in the video can be quickly captured, the calculation is simple, and the method is suitable for real-time processing.

The application discloses a pig farm mouse detection method, system, device and medium based on deep learning, and the method comprises the steps of: frame extraction processing is performed on a video of a to-be-detected area; image preprocessing is performed on an image frame set; a current frame after preprocessing is input into a mouse detection model to learn the appearance of a mouse; the mouse detection model is obtained by using an improved RT-DETR detection model; a pixel with a value of 255 in a binary image obtained by using a frame difference method is taken as a motion pixel, and a corresponding detection result is reserved; a mouse detection result is recorded, and a mouse appearing time length is obtained by weighted summation of the number of mice recorded after each frame detection, so as to calculate a mouse activity index of the to-be-detected area. By adding a frame difference prior branch, the binary image obtained by using the frame difference method is added to the decoding part of the RT-DETR algorithm, and double thresholds are set when the detection result is reserved, so that the model focuses more on the motion area in detection, and the detection success rate of small and fast mice is improved.

The invention relates to a real-time image defogging and moving target detection combined system. The system comprises a first VDMAIP core used for receiving a first data stream and sending the first data stream to a memory for storage; the moving target detection IP core is used for receiving the first data stream output by the image defogging IP core and the first data stream read and output by the first VDMAIP core from the memory, and performing moving target detection by adopting a frame difference method to obtain a second data stream; and the second VDMAIP core is used for receiving the second data stream and sending the second data stream to the display for displaying. According to the invention, a module design is adopted to integrate image defogging and moving target detection functions into a complete system so as to realize accurate detection of moving objects in fog; meanwhile, the whole system is developed by means of a ZYNQ embedded platform, FPGA hardware resources of the system are fully utilized to accelerate algorithm processing, and the real-time performance of the system is effectively improved.

The application relates to a high-efficiency distributed video compressed sensing reconstruction system and method, and belongs to the technical field of image processing. ++ The ISTA-Net is introduced into the distributed video compressed sensing system to independently reconstruct and initially reconstruct key frames and non-key frames, and the intra-frame correlation is fully utilized. Through the designed two-stage deep reconstruction network module, the inter-frame correlation is fully utilized to deeply reconstruct the non-key frames. The deep reconstruction network module FMC-Net of the first stage carries out motion compensation to obtain the non-key frames of the preliminary enhanced reconstruction; the deep reconstruction network module FDR-Net of the second stage carries out residual reconstruction by using frame difference information to recover details to obtain the finally reconstructed non-key frames. The method can significantly improve the reconstruction quality of the non-key frames, and high-quality video fast reconstruction is realized.