A big data rapid processing operation method based on a smart city

Abstract

The application is a kind of big data fast processing operation method based on smart city, relates to computer processing technical field, the method includes the following steps: S01, the video data package collected in the preset time period is obtained to the city road camera, the picture content feature reflecting traffic dynamics in the video data package is extracted, wherein the picture content feature divides the coverage area into a plurality of data blocks; S02, the key frame in the video data package is extracted for each data block, the space-time similarity of adjacent key frames is calculated and combined with time series analysis, and the data redundancy degree evaluation result is generated; S03, the video data package is adaptively compressed according to the data redundancy degree evaluation result. The application constructs a space-time joint redundancy evaluation model, designs a heterogeneous computing pipeline and a dynamic storage scheduling algorithm, and realizes the improvement of processing efficiency and the reduction of storage cost under the premise of ensuring information integrity.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer processing, in particular to a big data rapid processing operation method based on smart city. BACKGROUND

[0002] With the acceleration of smart city construction process, the scale of urban road video monitoring network grows exponentially. According to statistics, a single camera produces video data of hundreds of GB per day, and the massive data produced by a million camera clusters pose a serious challenge to the existing processing system.

[0003] The traditional video data processing method mainly has the following defects:

[0004] 1) The use of full storage mode leads to a storage resource waste rate of more than 60%;

[0005] 2) Frame extraction processing based on fixed time interval is easy to lose key event information;

[0006] 3) General compression algorithm is difficult to take into account the traffic scene characteristics, resulting in low compression efficiency;

[0007] 4) The data processing flow and storage strategy lack dynamic coordination mechanism, and the system response delay is significant.

[0008] Therefore, it is urgent to develop a big data processing method that integrates intelligent partitioning, adaptive compression and dynamic storage coordination to achieve the coordinated optimization of processing efficiency and storage efficiency under the premise of ensuring data integrity. SUMMARY

[0009] In view of the above technical problems, the technical scheme adopted by the present application is a big data rapid processing operation method based on smart city, which comprises the following steps:

[0010] S01, acquiring video data packets collected by urban road cameras in a preset time period, extracting picture content features reflecting traffic dynamics in the video data packets, wherein the picture content features divide the coverage area into multiple data blocks;

[0011] S02, extracting key frames in the video data packet for each data block, calculating the spatio-temporal similarity of adjacent key frames and combining time series analysis to generate a data redundancy degree evaluation result;

[0012] S03, performing adaptive compression processing on the video data packet according to the data redundancy degree evaluation result to obtain a compressed video data packet containing a first key data segment containing vehicle trajectory, pedestrian behavior and traffic signs;

[0013] S04, using a multi-modal deep learning model to process the compressed video data packets in parallel, extract structured traffic element information and generate visual analysis results, and as a second key data segment;

[0014] S05, based on real-time data flow and processing throughput, dynamically configure the storage path and resource allocation strategy of the distributed storage node to store the compressed video data packets and the second key data segment corresponding to the timestamp of the first key data segment.

[0015] As preferred, the video data packet acquisition in step S01 includes:

[0016] S11, control the camera to collect video streams at variable time intervals through the edge computing node, and the time interval is dynamically adjusted according to the light intensity;

[0017] S12, using frame difference method to detect motion area of original video stream to segment video clips containing effective traffic events to form data packets;

[0018] S13, input the video clip data packet into the YOLOv5-based invalid frame filtering model to identify and eliminate all-black frames, still frames and repeated frames, and generate preprocessed data packet index.

[0019] As preferred, the step S01 includes dividing the coverage area into multiple data blocks:

[0020] S14, extract multi-dimensional feature vectors in the video data packet, including vehicle density gradient, pedestrian movement heat map, and traffic sign distribution matrix;

[0021] S15, apply the DBSCAN clustering algorithm with space-time constraints, and fuse GIS coordinate data and time period weight factor , wherein t is the traffic collection time length, T m is the midpoint of the traffic peak period, N e is the current period historical accident number, and L is the road section grade coefficient, to generate feature blocks with similar traffic patterns;

[0022] S16, establish a block dynamic update mechanism, and trigger block redivision when three consecutive feature vectors deviate from the clustering center threshold.

[0023] As preferred, the data redundancy evaluation in step S02 includes:

[0024] S21, using an improved ORB feature point extraction algorithm, multi-scale spatial pyramid feature matching is performed on the key frame;

[0025] S22, construct a spatio-temporal similarity calculation model, integrate the HSV histogram intersection ratio, the SIFT descriptor cosine similarity and the optical flow motion vector similarity;

[0026] S23, design an LSTM-based time series analysis module to predict the redundancy change trend of the next five frames and generate a dynamic redundancy evaluation coefficient matrix.

[0027] As preferred, the adaptive compression processing in step S3 includes:

[0028] S31, create a priority queue of the key frames, set a reservation weight based on vehicle license plate recognition confidence and pedestrian posture anomaly index;

[0029] S32, based on a deployable hybrid compression engine, enable H.265 inter-frame prediction coding for high redundancy blocks and AV1 intra-frame coding for low redundancy blocks;

[0030] S33, based on an embedded SSIM-based quality feedback loop, automatically switch the coding mode when the compressed structural similarity is less than 0.95.

[0031] As preferred, the parallel processing of the compressed video data packet in step S04 includes:

[0032] S41, construct a multi-task deep learning network, the backbone network uses The parallel output branches include:

[0033] Vehicle detection branch: realize vehicle type classification and trajectory prediction based on an improved CenterNet;

[0034] Pedestrian analysis branch: extract abnormal behavior through OpenPose skeleton extraction combined with a spatio-temporal graph convolution network;

[0035] Traffic sign recognition branch: apply attention mechanism to enhance small target detection capability;

[0036] S42, design a feature fusion module to perform cross-modal alignment of visual features, radar point cloud data and meteorological sensor data;

[0037] S43, generate a structured traffic situation map, including a dynamic lane-level traffic heat map and an event spatio-temporal distribution matrix.

[0038] As preferred, the cross-modal alignment in step S42 includes:

[0039] S421, perform spatio-temporal synchronization of the radar point cloud data based on Kalman filtering;

[0040] S422, the weather data obtained by the weather sensor is matched with the video data through a sliding window correlation coefficient;

[0041] S423, a cross-modal attention map is established, and the calculation formula is wherein Q v is a visual feature query vector, K r is a radar feature key vector, and d is a dimension scaling factor.

[0042] Preferably, the storage strategy in step S05 is dynamically adjusted, including:

[0043] S51, a storage cost factor is calculated in real time Energy consumption, wherein: alpha is the access delay, beta is the storage density, and gamma is the energy consumption penalty coefficient;

[0044] S5, based on the storage path optimization algorithm , automatically select local SSD, distributed object storage or edge cache according to the data hot and cold characteristics;

[0045] S53, a double-layer redundancy elimination mechanism is implemented, BloomFilter-based deduplication is performed at the edge node, and content signature-based global deduplication is implemented at the center cloud.

[0046] The present application has at least the following beneficial effects:

[0047] 1. By fusing multi-dimensional feature vectors such as vehicle density gradient and pedestrian heat map, and combining the DBSCAN clustering algorithm with the space-time constraint, intelligent dynamic partitioning of the traffic scene is realized. Compared with the traditional fixed partitioning method, when the traffic flow changes, the block re-partitioning can be automatically triggered, so that the computing resources can accurately match the changes in the characteristics of each region, and the problem of resource waste caused by poor scene adaptability of the traditional method can be effectively solved.

[0048] 2. An improved ORB feature point extraction and space-time similarity calculation model is used, which combines cross-dimensional analysis of HSV histogram, SIFT descriptor and optical flow motion vector, and combines with the LSTM prediction module to build a multi-dimensional redundancy evaluation system. Compared with the single pixel comparison method, the accuracy of redundant data identification is improved, and the mixed compression engine and double-layer deduplication mechanism are used to reduce the storage demand, and the completeness of the key information, i.e. license plate and abnormal behavior information, is higher.

[0049] 3. Through the cooperative control of the SSIM quality feedback loop and the priority queue, dynamic optimization of the compression parameters is realized. Under the same compression rate, the structural similarity index of the present scheme is improved by 0.12 compared with the traditional H.265 encoding, and the vehicle license plate recognition rate is also improved. Especially for low-light scenes, the edge node can adaptively switch the encoding mode, ensuring the integrity of the key frame information at night, and solving the problem of night data distortion caused by fixed encoding in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0051] Figure 1 A big data fast processing operation method flow chart based on smart city is provided for the first embodiment of the present application.

[0052] Figure 2 A flow chart of S01 is provided for the first embodiment of the present application.

[0053] Figure 3 A flow chart of S02 is provided for the first embodiment of the present application.

[0054] Figure 4 A flow chart of S03 is provided for the first embodiment of the present application.

[0055] Figure 5 A flow chart of S04 is provided for the first embodiment of the present application. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0057] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and in the above-described drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0058] Embodiment one

[0059] The present embodiment provides a smart city-based big data rapid processing operation method, which comprises the following steps, as shown in the figure: Figure 1

[0060] S01, acquiring video data packets collected by a city road camera in a preset time period, extracting picture content features reflecting traffic dynamics in the video data packets, wherein the picture content features divide the coverage area into multiple data blocks;

[0061] Specifically, in combination with the figure, the acquisition of the above-mentioned video data packet comprises: Figure 2

[0062] S11, controlling the camera to collect video streams at a variable time interval by an edge computing node, and the time interval is dynamically adjusted according to the light intensity;

[0063] S12, using frame difference method to detect the motion area of the original video stream to segment out the video segment containing the effective traffic event to form a data packet;

[0064] S13, inputting the video segment data packet into a YOLOv5-based invalid frame filtering model to identify and eliminate all-black frames, still frames and repeated frames, and generating a pre-processing data packet index.

[0065] Through the edge computing node, the system can intelligently control the video collection time interval of the camera. This interval is no longer fixed, but is dynamically adjusted according to the light intensity and other conditions of the actual scene. For example, in the daytime with sufficient light, the system can appropriately lengthen the collection interval to reduce the data volume; while at night or in insufficient light, the collection interval is shortened to ensure the picture clarity, thereby optimizing the resource utilization. And the way of "collecting video streams at a variable time interval" can effectively avoid the problems of resource waste (such as collecting a large amount of useless data) and loss of key information (such as too long collection interval when important events occur) that may be caused by traditional fixed time interval collection.​​

[0066] Secondly, the frame difference method is used to detect the motion area of the original video stream. By comparing the differences between consecutive frames, the motion area in the picture can be accurately identified. Based on the results of motion area detection, the video segments containing effective traffic events can be automatically segmented. These segments usually contain important traffic information, such as vehicle driving, pedestrian crossing, etc.

[0067] Furthermore, the YOLOv5 model is used to intelligently analyze the video frames, which can accurately identify and eliminate invalid frames such as all-black frames, static frames, and repeated frames. These invalid frames usually do not contain useful traffic information, and eliminating them can significantly reduce the data volume for subsequent processing. After eliminating invalid frames, the system generates a pre-processing data package index. This index records the location and length of the effective video segments, providing convenience for subsequent processing.

[0068] Further, the method of dividing the coverage area into multiple data blocks in the above embodiment includes:

[0069] S14, extracting a multi-dimensional feature vector in the video data package, the multi-dimensional feature vector including a vehicle density gradient, a pedestrian movement heat map, and a traffic sign distribution matrix;

[0070] S15, applying a DBSCAN clustering algorithm with spatiotemporal constraints and fusing GIS coordinate data and time period weight factors , wherein t is the length of the traffic collection time, T m is the midpoint of the traffic peak period, N e is the historical accident number of the current period, and L is the road segment grade coefficient, to generate feature blocks with similar traffic patterns;

[0071] S16, establishing a block dynamic update mechanism, which triggers block redivision when the deviation of three consecutive feature vectors from the clustering center threshold is triggered.

[0072] The above extracts multi-dimensional feature vectors from the video data package, which contains rich information of the video content, such as vehicle density and pedestrian movement trajectory. Through the extracted feature vectors, the system applies a DBSCAN clustering algorithm with spatiotemporal constraints to divide the video picture into multiple feature blocks with similar traffic patterns, which helps the system to process different regions according to their characteristics and improve processing efficiency. For example, one area may be mainly straight vehicles, while another area may be mainly turning vehicles. The system can divide these areas into different feature blocks through feature vector extraction and clustering algorithm, and optimize the processing of each block according to its characteristics.

[0073] Secondly, when the feature vectors of three consecutive frames deviate from the cluster center threshold, the block re-partitioning is triggered to ensure that the system can adapt to the changes in the traffic scene in real time and maintain the efficiency of processing. By dynamically updating the block partitioning, the system can continuously maintain efficient and accurate processing of the traffic scene, even in the case of significant changes in traffic patterns.

[0074] S02, extract the key frames in the video data packet for each data block, calculate the spatio-temporal similarity of adjacent key frames, and generate a data redundancy degree evaluation result by combining time series analysis;

[0075] Specifically, in combination with Figure 3 As shown in the above data redundancy evaluation method includes:

[0076] S21, using an improved ORB feature point extraction algorithm, multi-scale spatial pyramid feature matching is performed on the key frames;

[0077] S22, construct a spatio-temporal similarity calculation model, integrate HSV histogram intersection ratio, SIFT descriptor cosine similarity and optical flow motion vector similarity;

[0078] S23, design a time series analysis module based on LSTM, predict the redundancy change trend of the next five frames, and generate a dynamic redundancy evaluation coefficient matrix.

[0079] The above accurately extracts key frames from video data packets. These key frames usually contain important information of the video content. By calculating the spatio-temporal similarity of adjacent key frames, the system can accurately evaluate the redundancy degree of video data. This evaluation provides a scientific basis for subsequent compression processing, which helps to reduce the data volume while ensuring video quality. For example, if the content similarity of consecutive frames in a video is very high, the system can identify this feature through spatio-temporal similarity calculation. In this way, in subsequent compression processing, the system can take appropriate optimization measures, such as increasing the compression ratio or skipping the processing of certain frames, to further improve processing efficiency.

[0080] Further, the generation of the above-mentioned time series analysis module based on LSTM includes:

[0081] Data input:

[0082] The time series analysis module based on LSTM receives feature vectors of video data as input. These feature vectors may include visual features such as brightness, color, texture of video frames, and processing features such as compression ratio, encoding method of video data.

[0083] Network training:

[0084] During the training phase, the LSTM network learns patterns of redundancy variation using a large amount of historical video data. The network adjusts its internal parameters based on the input feature vector and the corresponding redundancy label (e.g., high redundancy, low redundancy) to minimize prediction error.

[0085] Trend Forecast:

[0086] During the prediction phase, the LSTM network receives the feature vector of the current video frame and, based on its internal state and learned patterns, predicts the redundancy trend for the next five frames. The prediction results are output as a redundancy trend curve, visually demonstrating how redundancy changes over time.

[0087] Generation of dynamic redundancy evaluation coefficient matrix

[0088] Redundancy quantification:

[0089] Based on the redundancy change trend predicted by the LSTM network, the system quantifies redundancy into specific numerical values. These values ​​reflect the degree of redundancy of video data at different time points; higher values ​​indicate greater redundancy.

[0090] Matrix construction:

[0091] The system arranges the quantified redundancy values ​​in chronological order to construct a dynamic redundancy evaluation coefficient matrix. The rows of this matrix represent different time points (frames), and the columns represent the dimensions of redundancy evaluation (such as visual feature redundancy, processing feature redundancy, etc.).

[0092] Dynamic adjustment:

[0093] As video data is continuously input and processed, the dynamic redundancy assessment coefficient matrix is ​​updated in real time. The system adjusts the video data processing strategy, such as compression ratio and encoding method, based on the latest redundancy assessment results to optimize processing efficiency and quality.

[0094] Therefore, when processing a traffic monitoring video, the LSTM-based time series analysis module predicts that the redundancy of the next five frames will first increase and then decrease. Based on this prediction, the system can dynamically adjust the compression strategy: during the phase of increasing redundancy, a more efficient compression algorithm is used to reduce the amount of data; during the phase of decreasing redundancy, the compression ratio is appropriately reduced to retain more video details. Simultaneously, the system will output the redundancy evaluation results in matrix form for subsequent processing and analysis.

[0095] S03. Based on the data redundancy assessment results, the video data packet is adaptively compressed to obtain a compressed video data packet containing the first key data segment containing vehicle trajectory, pedestrian behavior and traffic signs.

[0096] Specifically, in combinationFigure 4 As shown, the adaptive compression process includes:

[0097] S31, create a priority queue of key frames, set a retention weight based on vehicle license plate recognition confidence and pedestrian posture anomaly index;

[0098] S32, based on the deployment of a configurable hybrid compression engine, enable H.265 inter-frame prediction coding for high redundancy blocks, and use AV1 intra-frame coding for low redundancy blocks;

[0099] S33, based on the embedded SSIM-based quality feedback loop, automatically switch the coding mode when the compressed structural similarity is less than 0.95.

[0100] During operation, the system creates a priority queue of key frames, sets a retention weight based on vehicle license plate recognition confidence, pedestrian posture anomaly index, and other factors. This mechanism ensures that critical information is not lost during compression. Then a hybrid compression engine is used to perform adaptive compression processing on video data packets based on data redundancy evaluation results. For high redundancy blocks, the engine uses efficient compression algorithms (such as H.265 inter-frame prediction coding) for compression; for low redundancy blocks, it uses compression algorithms that preserve details (such as AV1 intra-frame coding). Finally, an SSIM quality feedback loop is embedded, and when the compressed structural similarity is below the threshold, the coding mode will automatically switch to ensure that the compressed video quality meets the application requirements. That is, when processing a video containing vehicle driving and pedestrian crossing, the system can identify key information such as vehicle license plates and pedestrian postures, and through a priority queue, ensure that these information is not lost during compression. At the same time, for high redundancy areas in the video (such as static backgrounds), the system uses efficient compression algorithms for compression; while for low redundancy areas containing key information, it uses compression algorithms that preserve details.

[0101] S04 uses a multi-modal deep learning model to perform parallel processing on compressed video data packets, extracts structured traffic element information and generates visual analysis results, and serves as the second key data segment;

[0102] Further, in combination with Figure 5 As shown, using a multi-modal deep learning model to perform parallel processing on compressed video data packets includes:

[0103] S41, construct a multi-task deep learning network, the backbone network uses Parallel output branches include:

[0104] Vehicle detection branch: vehicle type classification and trajectory prediction based on improved CenterNet;

[0105] Pedestrian analysis branch: abnormal behavior is identified through OpenPose skeleton extraction combined with spatio-temporal graph convolution network;

[0106] Traffic sign recognition branch: attention mechanism is applied to enhance small target detection capability;

[0107] S42, a feature fusion module is designed to align the visual features with the radar point cloud data and the meteorological sensor data in a cross-modal manner;

[0108] S43, a structured traffic situation atlas is generated, including a dynamic lane-level flow heat map and an event spatio-temporal distribution matrix.

[0109] Secondly, the above cross-modal alignment includes:

[0110] S421, Kalman filter-based spatio-temporal synchronization is performed on the radar point cloud data;

[0111] S422, the meteorological data obtained by the meteorological sensor data and the video data are matched through a sliding window correlation coefficient;

[0112] S423, a cross-modal attention map is established, and the calculation formula is , wherein Q v is a visual feature query vector, K r is a radar feature key vector, and d is a dimension scaling factor.

[0113] Specifically, a multi-task deep learning network is built, taking as the backbone network, and multiple tasks such as vehicle detection, pedestrian analysis, and traffic sign recognition are processed in parallel. Through the feature fusion module, the visual features are aligned with the radar point cloud data and the meteorological sensor data in a cross-modal manner, improving the accuracy and robustness of detection. Finally, a structured traffic situation atlas is generated, including a dynamic lane-level flow heat map and an event spatio-temporal distribution matrix, providing strong support for traffic management.

[0114] Secondly, through Kalman filter synchronization of radar point cloud and video data, and sliding window correlation coefficient matching of meteorological information, the spatio-temporal alignment error of multi-source data is reduced. In the rain and snow weather test, the trajectory prediction error is reduced. The problem of "spatio-temporal asynchrony" caused by the difference in sampling frequency and coordinate system of multi-sensor data is solved. For example, the high-precision ranging of radar point cloud and the low-delay visual information of video are fused through the attention map, improving the robustness of target recognition in complex scenes.

[0115] S05, based on real-time data flow and processing throughput, dynamically configure the storage path and resource allocation strategy of the distributed storage node to store the compressed video data packet and the second key data segment corresponding to the timestamp of the first key data segment.

[0116] The storage strategy dynamic adjustment includes:

[0117] S51, real-time calculation of storage cost factor Energy consumption, wherein: alpha is the access delay (value range: 0.8-1.2), beta is the storage density (value range: 0.5-0.9), and gamma is the energy consumption penalty coefficient (value range: 0.3-0.7);

[0118] S5, based on The storage path optimization algorithm automatically selects local SSD, distributed object storage or edge cache according to the data hot and cold characteristics;

[0119] S53, a double-layer redundancy elimination mechanism is implemented, BloomFilter-based deduplication is performed at the edge node, and content signature-based global deduplication is implemented at the center cloud.

[0120] Specifically, the storage cost factor is calculated in real time to guide resource allocation. At the same time, a storage path optimization algorithm based on The storage path optimization algorithm automatically selects local SSD, distributed object storage or edge cache according to the data hot and cold characteristics. In addition, a double-layer redundancy elimination mechanism is implemented, BloomFilter-based deduplication is performed at the edge node, and content signature-based global deduplication is implemented at the center cloud, further reducing the storage cost and improving the access performance.

[0121] The first embodiment fuses multi-dimensional feature vectors such as vehicle density gradient and pedestrian heat map, and combines the DBSCAN clustering algorithm with the space-time constraint to realize intelligent dynamic partitioning of the traffic scene. Compared with the traditional fixed partitioning method, the block re-partitioning can be automatically triggered when the traffic flow changes, so that the computing resources can accurately match the changes in the characteristics of each region, effectively solving the problem of resource waste caused by poor scene adaptability of the traditional method. An improved ORB feature point extraction and space-time similarity calculation model is used, which integrates cross-dimensional analysis of HSV histogram, SIFT descriptor and optical flow motion vector, and combines with the LSTM prediction module to build a multi-dimensional redundancy evaluation system. Compared with the single pixel comparison method, the accuracy of redundant data recognition is improved, and the storage demand is reduced by cooperating with the hybrid compression engine and the double-layer deduplication mechanism, and the completeness of the key information, i.e. the license plate and abnormal behavior information, is higher. Secondly, through the cooperative control of the SSIM quality feedback loop and the priority queue, the dynamic optimization of the compression parameters is realized. Under the same compression rate, the structural similarity index of the present scheme is improved by 0.12 compared with the traditional H.265 encoding, and the vehicle license plate recognition rate is improved. Especially for low-light scenes, the edge node can adaptively switch the encoding mode to ensure the integrity of the key frame information at night, solving the problem of night data distortion caused by fixed encoding in the prior art.

[0122] Example two

[0123] The embodiment of the present application provides a kind of non-transient computer readable storage medium, at least one instruction or at least one program is stored in the non-transient computer readable storage medium, at least one instruction or at least one program is loaded and executed by processor to realize the steps of:

[0124] Acquire the video data package collected in preset time period by city road camera, extract the picture content features reflecting traffic dynamic in video data package, wherein picture content features will be divided into multiple data blocks by coverage area;

[0125] Extract the key frame in video data package for each data block, calculate the space-time similarity of adjacent key frame and combine time series analysis, generate data redundancy degree evaluation result;

[0126] According to data redundancy degree evaluation result, video data package is adaptively compressed to obtain compressed video data package containing the first key data section including vehicle trajectory, pedestrian behavior and traffic sign;

[0127] Multi-modal deep learning model is used to carry out parallel processing to compressed video data package, extract structured traffic element information and generate visual analysis result, and as the second key data section;

[0128] Based on real-time data flow and processing throughput, the storage path and resource allocation strategy of distributed storage node are dynamically configured to store compressed video data package and the second key data section corresponding to the timestamp of the first key data section.

[0129] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing relevant hardware, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, the processes of the above-mentioned embodiments of each method can be included. Any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0130] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the above-described functions.

[0131] Embodiment three

[0132] The embodiment of the application provides an electronic device, including a processor and a memory, at least one instruction or at least one program is stored in the memory, the at least one instruction or the at least one program is loaded and executed by the processor to realize the steps of:

[0133] Obtaining video data packets collected by a city road camera in a preset time period, extracting picture content features reflecting traffic dynamics in the video data packets, wherein the picture content features divide the coverage area into a plurality of data blocks;

[0134] Extracting key frames in the video data packets for each data block, calculating the spatio-temporal similarity of adjacent key frames and combining time series analysis to generate a data redundancy degree evaluation result;

[0135] According to the data redundancy degree evaluation result, the video data packet is adaptively compressed to obtain a compressed video data packet containing a first key data segment of vehicle trajectory, pedestrian behavior and traffic sign;

[0136] A multi-modal deep learning model is used to perform parallel processing on the compressed video data packet, extract structured traffic element information and generate visual analysis results, and serve as a second key data segment;

[0137] Based on real-time data flow and processing throughput, the storage path and resource allocation strategy of the distributed storage node are dynamically configured to store the compressed video data packet and the second key data segment corresponding to the timestamp of the first key data segment.

[0138] The above is only a preferred embodiment of the present application, not any form of limitation on the present application, although the present application has been disclosed as above with a preferred embodiment, however, it is not intended to limit the present application, any person skilled in the art, without departing from the technical solution of the present application, can make some changes or modifications to the above disclosed technical content to make equivalent embodiments with equivalent changes, but any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application, all still belong to the scope of the technical solution of the present application.

Claims

1. A method for smart city-based big data fast processing operation, characterized in that, The method comprises the following steps: S01, acquiring video data packets collected by urban road cameras within a preset time period, and extracting picture content features reflecting traffic dynamics in the video data packets, wherein the picture content features divide the coverage area into multiple data blocks; S02, extracting key frames in the video data packets for each data block, calculating the spatio-temporal similarity of adjacent key frames, and generating a data redundancy degree evaluation result in combination with time series analysis; S03, performing adaptive compression processing on the video data packets according to the data redundancy degree evaluation result to obtain compressed video data packets containing a first key data segment including vehicle trajectories, pedestrian behaviors and traffic signs; S04, performing parallel processing on the compressed video data packets using a multi-modal deep learning model, extracting structured traffic element information, and generating a visual analysis result as a second key data segment; S05, dynamically configuring the storage path and resource allocation strategy of the distributed storage node based on real-time data flow and processing throughput to store the compressed video data packets and the second key data segment corresponding to the timestamp of the first key data segment. 2.The big data fast processing operation method based on smart city according to claim 1, wherein, The acquisition of the video data packets in step S01 comprises: S11, controlling the camera to collect video streams at a variable time interval through an edge computing node, wherein the time interval is dynamically adjusted according to the light intensity; S12, performing motion region detection on the original video stream using frame difference method to segment out video segments containing effective traffic events to form data packets; S13, inputting the video segments into a YOLOv5-based invalid frame filtering model to identify and remove all-black frames, still frames and duplicate frames, and generating a preprocessed data packet index. 3.The big data fast processing operation method based on smart city according to claim 1, wherein, The division of the coverage area into multiple data blocks in step S01 comprises: S14, extracting a multi-dimensional feature vector in the video data packet, wherein the multi-dimensional feature vector includes vehicle density gradient, pedestrian movement heat map and traffic sign distribution matrix; S15, DBSCAN clustering algorithm with space-time constraints is applied, and GIS coordinate data and time period weight factor are fused wherein t is the length of traffic collection time, T m is the midpoint of the traffic peak period, N e is the historical accident number of the current period, and L is the road section grade coefficient, so as to generate a feature block with similar traffic mode; S16, establishing a block dynamic update mechanism, and triggering block redivision when three consecutive feature vectors deviate from the clustering center threshold. 4.The big data fast processing operation method based on smart city according to claim 1, wherein, The data redundancy evaluation in step S02 comprises: S21, using an improved ORB feature point extraction algorithm to perform multi-scale spatial pyramid feature matching on the key frames; S22, constructing a spatio-temporal similarity calculation model to comprehensively consider the HSV histogram intersection ratio, SIFT descriptor cosine similarity and optical flow motion vector similarity; S23, designing a time series analysis module based on LSTM to predict the redundancy change trend of the next five frames and generate a dynamic redundancy evaluation coefficient matrix. 5.The big data fast processing operation method based on smart city according to claim 1, wherein, The adaptive compression processing in step S3 comprises: S31, creating a priority queue of the key frames, and setting a reservation weight based on vehicle license plate recognition confidence and pedestrian posture anomaly index; S32, based on a deployable hybrid compression engine, enabling H.265 inter-frame prediction coding for high redundancy blocks, and using AV1 intra-frame coding for low redundancy blocks; S33, based on embedding the SSIM-based quality feedback loop, when the structural similarity after compression is less than 0.95, automatically switch the encoding mode. 6.The big data fast processing operation method based on smart city according to claim 1, wherein, The step S04 includes: S41, construct a multi-task deep learning network, the backbone network adopts The parallel output branch includes: Vehicle detection branch: based on the improved CenterNet to realize vehicle classification and trajectory prediction; Pedestrian analysis branch: through OpenPose skeleton extraction combined with spatio-temporal graph convolution network to identify abnormal behavior; Traffic sign recognition branch: application of attention mechanism to enhance small target detection ability; S42, design feature fusion module, cross-modal alignment of visual features, radar point cloud data and meteorological sensor data; S43, generate structured traffic situation atlas, including dynamic lane-level traffic heat map and event spatio-temporal distribution matrix. 7.The big data fast processing operation method based on smart city according to claim 6, wherein, The step S42 includes: S421, Kalman filter-based spatio-temporal synchronization is performed on the radar point cloud data; S422, the meteorological data obtained by the meteorological sensor data is matched with the video data through sliding window correlation coefficient; S423、establish a cross-modal attention map, the calculation formula is wherein Q v is a visual feature query vector, K r is a radar feature key vector, and d is a dimension scaling factor. 8.The big data fast processing operation method based on smart city according to claim 1, wherein, The step S05 includes: S51, calculating a storage cost factor in real time energy consumption, wherein: a is access latency, β is storage density, and γ is an energy consumption penalty coefficient; S52、based on a storage path optimization algorithm that automatically selects local SSD, distributed object storage, or edge cache based on data hotness characteristics; S53, implement a double-layer redundancy elimination mechanism, perform BloomFilter-based deduplication on the edge node, and implement content signature-based global deduplication on the center cloud. 9.A non-transitory computer-readable storage medium having stored therein at least one instruction or at least one piece of program, characterized in that, The at least one instruction or the at least one program is loaded and executed by the processor to realize the steps of the big data fast processing operation method based on smart city as claimed in any one of claims 1-8.

10. An electronic device, comprising: The processor and the memory, the memory has at least one instruction or at least one program, the at least one instruction or the at least one program is loaded and executed by the processor to realize the steps of the big data fast processing operation method based on smart city as claimed in any one of claims 1-8.

Images