A traffic image low-bandwidth network transmission compression encryption system and method

Abstract

The application discloses a traffic image low-bandwidth network transmission compression encryption system and method, relates to the technical field of traffic information transmission control, and comprises a traffic image acquisition module, which is used for acquiring continuous traffic image data and organizing time sequence traffic image data in accordance with the time sequence of acquisition. In the application, a structure region division processing is performed on the time sequence traffic image data through a semantic structure separation module, the traffic image is divided into a semantic structure region and a background region, and semantic structure expression data and background texture expression data are respectively constructed in a hierarchical expression construction module. In the semantic structure expression construction process, a skeleton fidelity adjustment coefficient is introduced, a fidelity reconstruction processing is performed on the skeleton geometric description, the spatial consistency of the structure region is kept unchanged in the compression expression process, the corresponding relationship between the structure region and the semantic category is kept unchanged, the structure region boundary definition and the background texture continuity are kept at the same time.

Description

Technical Field

[0001] This invention relates to the field of traffic information transmission and control technology, and in particular to a low-bandwidth network transmission compression and encryption system and method for traffic images. Background Technology

[0002] With the continuous development of intelligent transportation systems and the widespread deployment of road monitoring equipment, traffic image acquisition devices are widely used in urban roads, highways, tunnels, and transportation hubs. Traffic image data is typically acquired in the form of consecutive frames and organized in chronological order to form time-series traffic image data, which is used for business processing such as traffic condition analysis, violation identification, and event detection.

[0003] In real-world deployment environments, traffic image acquisition devices are often located at edge or remote nodes. Affected by bandwidth limitations, link congestion, latency, and packet loss fluctuations, real-time transmission of continuous traffic images under low-bandwidth networks is prone to increased retransmissions and acknowledgment delays, impacting data integrity and timeliness. Existing solutions often employ uniform frame compression or simple bitrate control, which struggles to adapt to dynamic network changes. Traffic images contain semantic structure regions and background texture regions; existing compression methods lack layered differentiated processing, easily resulting in the loss of critical structural details when the compression ratio is increased. Simultaneously, frame-wide encryption increases overhead, and fixed compression-encryption ratios are difficult to match network fluctuations. Existing bitrate adjustment is based on only a single metric, lacking multi-dimensional correlation analysis and temporal modeling, leading to unreasonable sub-stream allocation. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a low-bandwidth network transmission compression and encryption system and method for traffic images.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a low-bandwidth network transmission compression and encryption system for traffic images, comprising: a traffic image acquisition module, used to acquire continuous traffic image data and organize it into time-series traffic image data according to the acquisition time sequence; a semantic structure separation module, used to perform semantic target parsing processing on the time-series traffic image data, divide the time-series traffic image data into structural regions according to semantic category information, and generate semantic structure separation data; and a layer expression construction module, used to construct semantic structure expression data and background texture expression data based on the semantic structure separation data, and organize the semantic structure expression data and background texture expression data into layered image expression data. The expression ratio control module generates expression ratio control data based on network state sampling data, and adjusts the ratio of semantic structure sub-code stream data and background texture sub-code stream data in subsequent layered compressed image data according to the expression ratio control data. The dual-channel compression generation module performs independent compression processing on the layered image representation data to generate semantic structure sub-codestream data and background texture sub-codestream data, and performs encryption processing on the semantic structure sub-codestream data to form layered compressed image data; The layered reconstruction module performs sub-stream parsing on the layered compressed image data, generates sub-stream reconstruction status information, and generates semantic structure expression data and background texture expression data respectively. Based on the semantic structure expression data and background texture expression data, it generates reconstructed traffic image data. The network status sampling module collects transmission confirmation information and fragment retransmission information during the transmission of layered compressed image data, and receives sub-stream reconstruction status information generated by the layered reconstruction module. It then organizes the transmission confirmation information, fragment retransmission information, and sub-stream reconstruction status information to generate network status sampling data.

[0006] As a further description of the above technical solution: The traffic image acquisition module timestamps the acquired traffic image data according to the acquisition time sequence, generating image unit data with acquisition time information; it performs time continuity verification processing on the image unit data, marking image unit data with missing or abnormal time jumps, and rearranging the image unit data according to the acquisition time sequence to generate time-aligned image data; it performs inter-frame consistency processing on the time-aligned image data, merging traffic image data corresponding to duplicate acquisition times, and retaining the unique image unit data corresponding to the acquisition time; the image unit data arranged in acquisition time sequence and processed with time continuity verification and inter-frame consistency processing are organized into time-series traffic image data.

[0007] As a further description of the above technical solution: The semantic structure separation module reads the temporal traffic image data frame by frame, extracts the pixel data in each frame, and establishes a corresponding frame identifier. Semantic category determination is performed on the pixel data in each frame to generate semantic category information corresponding to the pixel location. Based on the semantic category information, the pixel locations are clustered to form multiple structural regions, and the corresponding semantic category information is associated with each structural region to generate structural region partitioning results. Boundary extraction processing is performed on the structural region partitioning results to generate region boundary information for each structural region. The semantic category information, structural region partitioning results, and region boundary information are organized according to the frame identifier to form semantic structure separation data.

[0008] As a further description of the above technical solution: The hierarchical representation construction module extracts structural region division results, semantic category information, and region boundary information based on semantic structure separation data to generate a set of structural regions. It then performs morphological skeleton extraction processing on each region of the structural region set to generate a skeleton pixel set, and generates skeleton topological and geometric descriptions on the skeleton pixel set. Finally, a skeleton fidelity adjustment coefficient is introduced. Based on the skeleton fidelity adjustment coefficient The skeleton geometry description undergoes fidelity reconstruction, simplifying or refining the segmented sequences to generate a fidelity skeleton geometry description. Region boundary information, skeleton topology, fidelity skeleton geometry description, and semantic category information are organized by region number to form semantic structure representation data. A background region set is generated based on the structure region set, and texture statistics are performed on the background region set to generate a basic background texture representation. Residual extraction is performed on the structure region coverage locations to generate residual representations of the structure regions. A texture co-occurrence fusion coefficient is introduced. Based on the texture co-occurrence fusion coefficient The background texture base representation and the structural region residual representation are weighted and fused to generate a fused texture representation, which is then organized into background texture representation data. The weight of the background texture base representation is... The weights of the residual representation in the structural region are Semantic structure representation data and background texture representation data are aligned according to the acquisition time sequence of temporal traffic image data, a region number correspondence is established, and layered image representation data is organized.

[0009] As a further description of the above technical solution: The expression ratio control module organizes the transmission confirmation information, fragment retransmission information, and sub-stream reconstruction status information in the network status sampling data according to the acquisition time sequence to form the current time window network statistical sequence group; it then performs time-series memory convergence processing on the current time window network statistical sequence group and the previous time window network statistical sequence group to generate a memory network statistical sequence group, and introduces time-series memory coefficients. The temporal memory aggregation processing is based on the temporal memory coefficient. The retention rate of the network statistical sequence group from the previous time window is adjusted. Symbolization and permutation pattern statistical processing are performed on the memory network statistical sequence group to generate transmission acknowledgment permutation entropy, fragment retransmission permutation entropy, and reconstructed state permutation entropy. Time-window normalization is then applied to these three types of permutation entropy. A network state vector sequence is constructed based on the memory network statistical sequence group to generate recurrence rate and determinism. Time-window normalization is then applied to the recurrence rate and determinism. Coupling strength adjustment is then performed on the normalized permutation entropy, normalized recurrence rate, and determinism to generate congestion driving forces and introduce a coupling strength coefficient. The coupling strength adjustment process is based on the coupling strength coefficient. Adjust the fusion strength of permutation entropy, recurrence rate, and determinism; generate proportional relationship values ​​in expression ratio control data based on congestion driving quantity, the proportional relationship values ​​include the proportional values ​​corresponding to semantic structure sub-codestream data and background texture sub-codestream data, and satisfy that the sum of the two is 1; after generating expression ratio control data, adjust the generation ratio of semantic structure sub-codestream data and background texture sub-codestream data according to the proportional relationship values, so that the proportional relationship between semantic structure sub-codestream data and background texture sub-codestream data in subsequent layered compressed image data is consistent with the proportional relationship values.

[0010] As a further description of the above technical solution: The dual-channel compression generation module divides the semantic structure representation data and background texture representation data in the layered image representation data into data blocks according to the acquisition time sequence, generating a set of semantic structure data blocks and a set of background texture data blocks. It then performs independent compression processing on the semantic structure data block set to generate semantic structure sub-stream data; it also performs independent compression processing on the background texture data block set to generate background texture sub-stream data; finally, it encrypts the semantic structure sub-stream data to generate encrypted semantic structure sub-stream data; based on the proportional relationship value in the representation ratio control data, it adjusts the ratio of the encrypted semantic structure sub-stream data and the background texture sub-stream data by controlling the allocation ratio of the semantic structure sub-stream data and the background texture sub-stream data in the number of data blocks, the length of data blocks, or the order in which data blocks are sent, generating adjusted semantic structure sub-stream data and background texture sub-stream data; finally, it organizes and encapsulates the adjusted semantic structure sub-stream data and background texture sub-stream data according to the acquisition time sequence to generate layered compressed image data.

[0011] As a further description of the above technical solution: The layered reconstruction module performs sub-stream parsing on the layered compressed image data, separating semantic structure sub-stream data and background texture sub-stream data. It then decrypts the semantic structure sub-stream data to generate decrypted semantic structure sub-stream data. Next, it decompresses the decrypted semantic structure sub-stream data to generate semantic structure representation data. Finally, it decompresses the background texture sub-stream data to generate background texture representation data. The semantic structure representation data and background texture representation data are spatially aligned according to the acquisition time sequence of the temporal traffic image data to restore the pixel positions corresponding to the structural region division results. After spatial alignment, the semantic structure representation data is overlaid onto the corresponding structural region positions of the background texture representation data to generate reconstructed traffic image data.

[0012] As a further description of the above technical solution: The network state sampling module records the transmission confirmation information corresponding to the layered compressed image data during transmission; it also records the fragment retransmission information corresponding to the layered compressed image data during transmission; the layered reconstruction module generates sub-stream reconstruction status information based on the layered compressed image data; the transmission confirmation information, fragment retransmission information, and sub-stream reconstruction status information are aligned and organized according to the acquisition time sequence of the layered compressed image data to generate network state sampling data; the network state sampling data is transmitted to the expression ratio control module, which generates expression ratio control data based on the network state sampling data, and adjusts the ratio relationship between the semantic structure sub-stream data and the background texture sub-stream data in subsequent layered compressed image data according to the ratio relationship value in the expression ratio control data.

[0013] As a further description of the above technical solution: The traffic image acquisition module generates temporal traffic image data, which is then transmitted to the semantic structure separation module. The semantic structure separation module transmits the generated semantic structure separation data to the hierarchical representation construction module. The hierarchical representation construction module organizes the generated semantic structure representation data and background texture representation data into hierarchical image representation data, which is then transmitted to the representation ratio control module and the dual-channel compression generation module. The network state sampling module generates network state sampling data based on the hierarchical compressed image data, which is then transmitted to the representation ratio control module. The representation ratio control module generates representation ratio control data based on the hierarchical image representation data and network state sampling data, which is then transmitted to the dual-channel compression generation module. The dual-channel compression generation module generates semantic structure sub-codestream data and background texture sub-codestream data based on the hierarchical image representation data and representation ratio control data, and organizes them into hierarchical compressed image data. This hierarchical compressed image data is then transmitted to the network state sampling module and the hierarchical reconstruction module. Finally, the hierarchical reconstruction module generates semantic structure representation data and background texture representation data based on the hierarchical compressed image data, and then generates reconstructed traffic image data based on these data.

[0014] As a further description of the above technical solution: A method for compression and encryption of traffic images for low-bandwidth network transmission includes the following steps: Acquire continuous traffic image data, and perform time stamping, time continuity verification, and inter-frame consistency processing on the traffic image data according to the acquisition time sequence to generate time-series traffic image data; Semantic target parsing processing is performed on time-series traffic image data to generate semantic category information, structural region division results and region boundary information, and organize them into semantic structure separation data; Based on semantic structure separation data, a set of structural regions is generated. Morphological skeleton extraction is performed on the set of structural regions to generate skeleton topological and geometric descriptions. A skeleton fidelity adjustment coefficient α is introduced to reconstruct the skeleton geometric descriptions with fidelity, generating semantic structure expression data. Simultaneously, a set of background regions is generated based on the set of structural regions. Texture statistical processing is performed on the set of background regions to generate basic background texture expression. Residual extraction is performed on the coverage positions of structural regions to generate structural region residual expression. A texture co-occurrence fusion coefficient β is introduced to perform weighted fusion processing on the basic background texture expression and the structural region residual expression to generate background texture expression data. Finally, the semantic structure expression data and the background texture expression data are aligned according to the acquisition time order to generate layered image expression data. Semantic structure representation data and background texture representation data in the layered image representation data are divided into data blocks and compressed independently to generate semantic structure sub-code stream data and background texture sub-code stream data. The semantic structure sub-code stream data is encrypted and organized into layered compressed image data. During the transmission of layered compressed image data, transmission confirmation information and fragment retransmission information are recorded. The layered reconstruction module generates sub-stream reconstruction status information based on the layered compressed image data. The transmission confirmation information, fragment retransmission information, and sub-stream reconstruction status information are then combined to generate network status sampling data. The network state sampling data is processed by time-series memory aggregation and a time-series memory coefficient γ is introduced to generate a memory network statistical sequence group. The memory network statistical sequence group is then processed by symbolization and permutation pattern statistics to generate the transmission acknowledgment permutation entropy, fragment retransmission permutation entropy, and reconstructed state permutation entropy, and the recurrence rate and determinism are generated. A coupling strength coefficient δ is introduced to couple and adjust the permutation entropy, recurrence rate, and determinism to generate congestion driving force. Based on the congestion-driven quantity, the proportional relationship value in the expression proportional control data is generated, and the proportional relationship between the semantic structure sub-code stream data and the background texture sub-code stream data in the subsequent layered compressed image data is adjusted according to the proportional relationship value. The layered compressed image data is parsed, decrypted, and decompressed to generate semantic structure representation data and background texture representation data. The semantic structure representation data is then overlaid onto the structural regions corresponding to the background texture representation data to generate reconstructed traffic image data.

[0015] The present invention has the following beneficial effects: 1. In this invention, the temporal traffic image data is first divided into semantic structure regions and background regions by a semantic structure separation module. Semantic structure representation data and background texture representation data are then constructed separately in a hierarchical representation construction module. During the semantic structure representation construction process, a skeleton fidelity adjustment coefficient is introduced to perform fidelity reconstruction of the skeleton geometry, ensuring that the structural regions maintain spatial consistency and semantic category correspondence during compression. During the background texture representation construction process, a texture co-occurrence fusion coefficient is introduced to perform weighted fusion of the background texture basic representation and the structural region residual representation, ensuring that the clarity of the structural region boundaries and the continuity of the background texture are maintained simultaneously. Through these processes, while improving the compression ratio, the morphological features and boundary details of key semantic structure regions in the traffic image are effectively preserved, enhancing the usability of the reconstructed traffic image in traffic target recognition, behavior analysis, and event detection scenarios.

[0016] 2. In this invention, a dual-channel compression generation module independently compresses semantic structure representation data and background texture representation data to generate semantic structure sub-stream data and background texture sub-stream data, and only performs encryption processing on the semantic structure sub-stream data. This method avoids the computational overhead of uniformly encrypting the entire frame of the image, allowing the encryption processing to focus on semantic structure information. This reduces the overall processing burden while ensuring the security of semantic structure information and improves real-time processing efficiency in low-bandwidth environments. A network state sampling module performs unified time alignment and processing on transmission confirmation information, fragment retransmission information, and sub-stream reconstruction state information to generate network state sampling data. The expression ratio control module performs time-series memory aggregation processing on the network state sampling data in the time window dimension, introduces time-series memory coefficients to construct a memory network statistical sequence group, and generates congestion driving quantities based on permutation entropy, recurrence rate, and determinism. This mechanism can perform correlation analysis on network transmission behavior and reconstruction behavior at a unified time scale, reflecting the fluctuation trend and structural stability of network state, avoiding the ratio distortion problem caused by adjusting based on a single indicator, and improving the accuracy of ratio control.

[0017] 3. In this invention, a proportional relationship value between semantic structure sub-stream data and background texture sub-stream data is generated based on congestion driving factors. In the dual-channel compression generation module, the proportion of sub-streams in the layered compressed image data is kept consistent with the proportional relationship value by adjusting the number of data blocks, data block length, or data block sending order. This mechanism can increase the proportion of background texture sub-streams when the network is stable and prioritize the transmission of semantic structure sub-streams when the network is congested, thereby achieving dynamic allocation of bandwidth resources and improving effective transmission efficiency in low-bandwidth network environments. Simultaneously with the transmission of layered compressed image data, transmission confirmation information and fragment retransmission information are recorded. This data, combined with the sub-stream reconstruction status information generated by the layered reconstruction module, forms network status sampling data, which is then fed back to the expression ratio control module for ratio adjustment, forming a closed-loop structure between transmission, sampling, adjustment, and reconstruction. This closed-loop mechanism enables the image representation ratio to be adjusted in real time according to changes in network status, improving the system's stable operation in complex network environments. By performing time stamping, time continuity verification, and inter-frame consistency processing on traffic image data, the temporal sequence of traffic image data is kept consistent throughout the entire process. Each module uses the acquisition time as the organizational benchmark, ensuring a temporal correspondence between semantic structure representation data, background texture representation data, sub-code stream data, and network status sampling data. This enhances the coherence of data flow transmission between modules and improves the overall reliability of the system. Attached Figure Description

[0018] Figure 1 This is a system architecture diagram of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Reference Figure 1 The present invention provides an embodiment of a low-bandwidth network transmission compression and encryption system for traffic images, comprising: a traffic image acquisition module for acquiring continuous traffic image data and organizing it into time-series traffic image data according to the acquisition time sequence; a semantic structure separation module for performing semantic target parsing processing on the time-series traffic image data, dividing the time-series traffic image data into structural regions based on semantic category information, and generating semantic structure separation data; and a layer expression construction module for constructing semantic structure expression data and background texture expression data based on the semantic structure separation data, and organizing the semantic structure expression data and background texture expression data into layered image expression data. The expression ratio control module generates expression ratio control data based on network state sampling data, and adjusts the ratio of semantic structure sub-code stream data and background texture sub-code stream data in subsequent layered compressed image data according to the expression ratio control data. The dual-channel compression generation module performs independent compression processing on the layered image representation data to generate semantic structure sub-codestream data and background texture sub-codestream data, and performs encryption processing on the semantic structure sub-codestream data to form layered compressed image data; The layered reconstruction module performs sub-stream parsing on the layered compressed image data, generates sub-stream reconstruction status information, and generates semantic structure expression data and background texture expression data respectively. Based on the semantic structure expression data and background texture expression data, it generates reconstructed traffic image data. The network status sampling module collects transmission confirmation information and fragment retransmission information during the transmission of layered compressed image data, and receives sub-stream reconstruction status information generated by the layered reconstruction module. It then organizes the transmission confirmation information, fragment retransmission information, and sub-stream reconstruction status information to generate network status sampling data.

[0021] In this embodiment, after the traffic image acquisition module acquires continuous traffic image data, it performs time organization processing on the traffic image data according to the acquisition time sequence to generate time-series traffic image data.

[0022] First, the traffic image acquisition module reads the acquired traffic image data frame by frame and writes a corresponding acquisition time stamp to each frame, generating image unit data with acquisition time information. Each image unit data contains the traffic image data itself and its corresponding acquisition time information.

[0023] Subsequently, a time continuity check is performed on the image unit data. The image unit data is traversed according to the acquisition time sequence, and the time intervals between adjacent acquisition times are compared. When a missing acquisition time or an abnormal jump in acquisition time is detected, the corresponding image unit data is marked, while the original image unit data is retained without deletion.

[0024] After completing the time continuity verification process, the image unit data is rearranged according to the acquisition time order to generate time-aligned image data. In the time-aligned image data, the image unit data is arranged in order from earliest to latest acquisition time, and the acquisition time order remains monotonically increasing.

[0025] Next, inter-frame consistency processing is performed on the time-aligned image data. For cases where traffic image data corresponds to duplicate acquisition times, the image unit data corresponding to the duplicate acquisition times are merged, and the unique image unit data corresponding to the acquisition time is retained, so that each acquisition time corresponds to only one image unit data record.

[0026] Finally, the image unit data, arranged in chronological order of acquisition and having undergone temporal continuity verification and inter-frame consistency processing, are organized according to their acquisition time to form temporal traffic image data. This generated temporal traffic image data serves as input data for the subsequent semantic structure separation module.

[0027] In this embodiment, after receiving the time-series traffic image data generated by the traffic image acquisition module, the semantic structure separation module performs structural region division processing on the time-series traffic image data frame by frame to generate semantic structure separation data.

[0028] First, the semantic structure separation module reads the time-series traffic image data frame by frame in chronological order of acquisition, and establishes a corresponding frame identifier for each frame. The frame identifier maintains a correspondence with the acquisition time information of the traffic image data for that frame, and is used for subsequent organization of semantic structure separation data.

[0029] Subsequently, pixel data is extracted from each frame of the image, and semantic category determination processing is performed on the pixel data. The semantic category determination processing generates corresponding semantic category information for each pixel location, so that each pixel location is associated with a semantic category information.

[0030] After obtaining semantic category information, pixel locations are clustered based on this information. Pixel locations with the same semantic category information and spatially adjacent locations are merged to form multiple structural regions. Each structural region is associated with corresponding semantic category information, thus generating the structural region partitioning result.

[0031] After generating the structural region segmentation results, boundary extraction processing is performed on the segmentation results. Boundary extraction processing is used to determine the region boundary information of each structural region in the image. The region boundary information records the spatial contour position of the structural regions, which is used by the subsequent hierarchical representation construction module to generate semantic structure representation data.

[0032] Finally, the semantic category information, structural region segmentation results, and region boundary information are organized according to frame identifiers. Using frame identifiers as units, the semantic category information, structural region segmentation results, and region boundary information corresponding to the same frame image are integrated and arranged in chronological order of acquisition time to form semantic structure separation data. This generated semantic structure separation data serves as the input data for the hierarchical representation construction module.

[0033] In this embodiment, after receiving the semantic structure separation data generated by the semantic structure separation module, the hierarchical representation construction module performs structural representation and background representation separation construction processing on the semantic structure separation data to generate hierarchical image representation data.

[0034] First, the hierarchical representation construction module extracts structural region segmentation results, semantic category information, and region boundary information from the semantic structure separation data. Based on the structural region segmentation results, the structural regions under the same frame identifier are aggregated to generate a structural region set. Each structural region in the structural region set is associated with corresponding semantic category information and region boundary information.

[0035] Subsequently, morphological skeleton extraction is performed region by region on the structural region set. For each structural region, its skeleton pixel set is extracted, and a skeleton topological description and a skeleton geometric description are established on the skeleton pixel set. The skeleton topological description is used to record the connection relationships between skeleton pixels, and the skeleton geometric description is used to record the spatial position of the skeleton pixels in the image.

[0036] After obtaining the skeleton geometry description, a skeleton fidelity adjustment coefficient is introduced. A high-fidelity reconstruction is performed on the skeleton geometry. This is based on the skeleton fidelity adjustment coefficient. The segmented sequence in the skeleton geometric description is simplified or refined to generate a skeleton-fidelity geometric description while maintaining the correspondence between region boundary information and semantic category information. Subsequently, the region boundary information, skeleton topological description, skeleton-fidelity geometric description, and semantic category information are organized according to region numbering to form semantic structure representation data.

[0037] Under the same frame identifier, the hierarchical representation construction module reads each structural region in the set of structural regions one by one, extracts the region boundary information, skeleton topology description and skeleton geometry description corresponding to the structural region, and maintains the one-to-one correspondence between the structural region numbers.

[0038] The spatial correspondence between the skeleton geometric description and the region boundary information is calculated. The skeleton pixel set is mapped to the spatial contour defined by the region boundary information, and the spatial distance distribution from the skeleton pixels to the region boundary contour is calculated to generate skeleton boundary deviation data. This skeleton boundary deviation data characterizes the degree of spatial consistency between the skeleton geometric description and the region boundary information. The formula for calculating skeleton boundary deviation is: ; Skeleton boundary deviation data The number of pixels in the skeleton pixel set. : No. The spatial coordinates of each skeleton pixel in the skeleton geometry description. The set of boundary pixels corresponding to the region boundary information. Euclidean distance calculation. Calculates the average minimum distance from the skeleton pixel set to the region boundary contour.

[0039] After obtaining the skeleton boundary deviation data, statistical processing is performed on the skeleton topology description. The number of connections between skeleton pixels and the number of skeleton branch structures are counted to generate skeleton structure complexity data. This skeleton structure complexity data reflects the level of detail in the skeleton representation of the current structural region. The formula for calculating skeleton structure complexity is: ; Skeletal structure complexity data The number of branch structures in the skeleton topology description. The sum of the lengths of all branch paths in the skeleton topology description. : The pixel area corresponding to the structural region. Used to measure the level of detail in the skeleton representation.

[0040] Subsequently, within the same structural region, the coverage ratio of the skeleton pixel set within the structural region pixel set is statistically analyzed to generate skeleton coverage ratio data. This skeleton coverage ratio data reflects the degree to which the skeleton's geometric description expresses the morphology of the structural region. The skeleton coverage ratio calculation formula is as follows: ; Skeleton coverage ratio data : Number of skeleton pixels Total number of pixels in the structural region.

[0041] After obtaining skeleton boundary deviation data, skeleton structural complexity data, and skeleton coverage ratio data, the hierarchical representation construction module performs normalization processing on the above data within the current frame identifier, ensuring that the data across different structural regions are within a unified measurement range. Based on the correspondence between the normalized skeleton boundary deviation data and the skeleton coverage ratio data, and combined with the skeleton structural complexity data, a skeleton fidelity adjustment coefficient is generated. The skeleton boundary deviation and skeleton coverage ratio are normalized. ; Subsequently, skeleton fidelity adjustment coefficients are generated: , Skeleton fidelity adjustment coefficient Normalized skeleton boundary deviation Normalized skeleton coverage ratio : Skeletal structure complexity : Minimum and maximum deviation values ​​within the current frame : The minimum and maximum coverage ratio within the current frame.

[0042] Skeleton fidelity adjustment coefficient The value of is limited to the range of 0 to 1. As the deviation of the skeleton boundary increases, The corresponding increase occurs when the skeleton coverage ratio is high and the skeleton structural complexity remains stable. The corresponding decrease. The generated It is bound to the corresponding structural region number and used as an adjustment parameter in the subsequent simplification or refinement of the segmented sequence of the skeleton geometry description, so as to maintain the spatial consistency of the structural region and the correspondence between semantic category information during the compression expression process.

[0043] After generating the semantic structure representation data, a background region set is generated based on the structural region set. The background region set includes pixel locations outside the structural region set. Texture statistical processing is then performed on the background region set to generate the basic background texture representation.

[0044] Next, residual extraction is performed on the areas covered by the structural regions to generate structural region residual representations. These structural region residual representations are used to record the texture differences between the structural regions and the background regions.

[0045] After obtaining the basic representation of the background texture and the residual representation of the structural region, a texture co-occurrence fusion coefficient is introduced. A weighted fusion process is performed on the basic representation of background texture and the residual representation of structural regions. This is based on the texture co-occurrence fusion coefficient. To give the background texture a basic expression Weights are assigned to the residual representation of the structural region. Weights are used to generate a fused texture representation and organize it into background texture representation data.

[0046] Under the same frame identifier, the hierarchical representation construction module reads the background texture basic representation and the structural region residual representation, and maintains the correspondence between the two types of representation data and the region number.

[0047] Subsequently, within the area covered by the structural region, pixel difference statistical processing is performed between the background texture base representation and the structural region residual representation to generate texture difference amplitude data. The texture difference amplitude data is used to characterize the degree of offset between the structural region residual representation and the background texture base representation. The formula for calculating the texture difference amplitude is as follows: ; Texture difference amplitude data : Number of pixels covered by the structural region : No. The basic representation value of the background texture at each pixel location. : No. The residual representation value of the structural region at each pixel location.

[0048] After obtaining the texture difference amplitude data, the concentration of texture difference amplitudes in spatial distribution is statistically analyzed to generate texture difference distribution density data. Texture difference distribution density data reflects the spatial distribution of residual expressions in structural regions. The formula for calculating texture difference distribution density is as follows: ; Texture difference distribution density data The number of pixels whose difference magnitude is higher than the average difference magnitude of the current frame. : Number of pixels covered by the structural region.

[0049] Subsequently, the embedding ratio of the structural region residual representation in the background texture base representation is statistically processed to generate residual embedding ratio data. This residual embedding ratio data reflects the coexistence relationship between the structural region texture and the background texture. The formula for calculating the residual embedding ratio is: ; Residual embedding ratio data : Residual expression values ​​of structural regions : Basic representation value of background texture.

[0050] After obtaining texture difference amplitude data, texture difference distribution density data, and residual embedding ratio data, the hierarchical representation construction module performs normalization processing on the above data within the current frame identifier, and generates texture co-occurrence fusion coefficients based on the correspondence between the normalized texture difference amplitude data and the residual embedding ratio data. Normalize the difference amplitude and embedding ratio: , ; Generate texture co-occurrence blending coefficient: , Texture co-occurrence blending coefficient Normalized texture difference amplitude, Normalized residual embedding ratio Texture difference distribution density, : The minimum and maximum values ​​of the difference amplitude in the current frame. : The minimum and maximum embedding ratio of the current frame.

[0051] In this embodiment, the texture co-occurrence fusion coefficient The value of is limited to the range of 0 to 1. When the texture difference amplitude increases and the residual embedding ratio increases, The corresponding increase occurs when the magnitude of the difference between the residual representation of the structural region and the basic representation of the background texture decreases. The corresponding decrease. The generated It is bound to the current frame identifier and participates in the weight allocation calculation when performing weighted fusion processing on the background texture base representation and the structural region residual representation, thereby maintaining the spatial continuity of the background texture representation while preserving the clarity of the structural region boundaries. Texture fusion expression formula: ; : Integrating texture representation, : Basic expression of background texture : Residual representation of structural regions Texture co-occurrence fusion coefficient.

[0052] Finally, the semantic structure representation data and background texture representation data are aligned according to the acquisition time sequence of the temporal traffic image data. Using frame identifiers as units, a region number correspondence is established between the semantic structure representation data and background texture representation data corresponding to the same frame, and these are arranged according to the acquisition time sequence to form layered image representation data. The generated layered image representation data serves as input data for the representation ratio control module and the dual-channel compression generation module.

[0053] In this embodiment, after receiving the network state sampling data generated by the network state sampling module, the expression ratio control module performs ratio control calculation processing on the network state sampling data to generate expression ratio control data.

[0054] First, the expression ratio control module reads transmission confirmation information, fragment retransmission information, and sub-stream reconstruction status information from the network status sampling data. These three types of information are then organized according to the acquisition time sequence to form the current time window network statistical sequence group.

[0055] Subsequently, the current time window network statistical sequence group is aligned with the previous time window network statistical sequence group, and time-series memory convergence processing is performed. A time-series memory coefficient is introduced. During the aggregation process, based on the time sequence memory coefficient The retention rate of the network statistical sequence group from the previous time window is adjusted to generate a memory network statistical sequence group. The memory network statistical sequence group reflects the combined statistical results of the network state sampling data from the current time window and the previous time window.

[0056] The expression ratio control module reads the network statistical sequence group of the current time window and the network statistical sequence group of the previous time window in the order of acquisition time. Both the current time window network statistical sequence group and the previous time window network statistical sequence group include transmission confirmation information, fragment retransmission information and sub-stream reconstruction status information, and maintain a one-to-one correspondence with the acquisition time.

[0057] Subsequently, at the same acquisition time location, differential statistical processing is performed on the data corresponding to the network statistical sequence group of the current time window and the network statistical sequence group of the previous time window. The differential results for transmission acknowledgment information, fragment retransmission information, and sub-stream reconstruction status information are calculated respectively. These differential results are then processed within the current time window to generate time window status difference data. The formula for calculating time window status difference is: ; : No. Data on the differences in state between time windows. : Statistics of confirmation messages sent during the current time window : Current time window segmented retransmission statistics : Current time window bitstream reconstruction status statistics. : Statistical value corresponding to the previous time window : Difference weighting coefficient, satisfying .

[0058] After obtaining the time window state difference data, sliding statistical processing is performed on the time window state difference data within multiple consecutive time windows to generate time window stability data. Time window stability data reflects the degree of continuous change in network state over time. The formula for calculating time window stability is: ; : No. Time window stability data for each time window : Number of sliding statistical time windows Historical time window status difference data.

[0059] Subsequently, the stability data for the time windows is normalized within the current statistical interval to ensure that the stability data across different time windows falls within a uniform measurement range. Time-series memory coefficients are then generated based on the normalized stability data for the time windows. . The value of is limited to the range of 0 to 1. Temporal memory coefficient Formula generation: First, normalize the stability: Generate temporal memory coefficients: ; : No. The temporal memory coefficient of each time window Normalized time window stability : The minimum and maximum values ​​of stability within the current statistical interval. .

[0060] In this embodiment, when the normalized time window stability data is large, A larger value increases the retention rate of the previous time window network statistical sequence group in the memory network statistical sequence group; when the normalized time window stability data is small... A smaller value gives the current time window network statistical sequence group a higher weight in the aggregation process. Time-series memory aggregation formula: ; : No. Statistical sequence set of memory networks within a time window : Current time window network statistical sequence group : The network statistical sequence group of the previous time window : Temporal memory coefficient.

[0061] In generation Subsequently, the expression ratio control module, when performing time-series memory convergence processing, according to... The network statistical sequence set from the previous time window is weighted and retained, and then merged with the network statistical sequence set from the current time window to generate a memory network statistical sequence set. This memory network statistical sequence set is used for subsequent permutation entropy calculation and network state vector sequence construction. Permutation entropy calculation formula; taking the sending of confirmation information as an example: ; Send confirmation of entropy. Number of permutation patterns : No. The probability of a certain permutation pattern occurring within a time window, and the permutation entropy during fragmented retransmission. Reconstructing the state arrangement entropy The calculation method is the same. Normalization: .

[0062] After generating the memory network statistical sequence set, symbolization and permutation pattern statistical processing are performed on the memory network statistical sequence set. A transmission acknowledgment permutation entropy is generated for the data corresponding to the transmission acknowledgment information, a fragment retransmission permutation entropy is generated for the data corresponding to the fragment retransmission information, and a reconstruction state permutation entropy is generated for the data corresponding to the sub-stream reconstruction state information. Subsequently, time-window normalization processing is performed on the transmission acknowledgment permutation entropy, fragment retransmission permutation entropy, and reconstruction state permutation entropy. Reproducibility and determinism calculation formulas: ; ; Recurrence rate Certainty : Length of the network state vector sequence : No. A network state vector, : Reproduce the distance threshold for judgment Indicator functions : Length is The number of diagonal segments, Minimum continuous length Maximum continuous length. Normalization: ; .

[0063] Next, a network state vector sequence is constructed based on the statistical sequence set of the memory network, and the recurrence rate and determinism are generated from this network state vector sequence. The recurrence rate and determinism are normalized within a time window to ensure that the recurrence rate and determinism are within the same measurement range as the permutation entropy.

[0064] After obtaining the normalized permutation entropy, normalized recurrence rate, and determinism, coupling strength adjustment is performed. A coupling strength coefficient is introduced. During the coupling process, based on the coupling strength coefficient By adjusting the fusion strength of permutation entropy, recurrence rate, and determinism, congestion driving forces are generated.

[0065] First, the expression ratio control module reads the normalized transmission confirmation arrangement entropy, fragment retransmission arrangement entropy, and reconstruction state arrangement entropy, and at the same time reads the normalized reproducibility and determinism, and maintains the one-to-one correspondence between the above data and the current time window.

[0066] Subsequently, a network state vector sequence is constructed based on the memory network statistical sequence group, and the change amplitude of the network state vector sequence within the current time window is statistically processed to generate network state fluctuation data.

[0067] After obtaining network state fluctuation data, correlation statistical processing is performed on the permutation entropy sequence, recurrence rate, and deterministic sequence to generate entropy-structure correlation data. Entropy-structure correlation data reflects the degree of correlation between permutation entropy changes and network state structural characteristics. The formula for calculating entropy-structure correlation is: ; Entropy—structural correlation data. : Permutation entropy sequence, Recurrence rate and deterministic combination sequences Mean entropy of permutations : Mean value of structural features.

[0068] Subsequently, within the current time window, network state fluctuation data and entropy-structure correlation data are jointly statistically processed to generate coupling sensitivity data. Coupling sensitivity data reflects the degree to which network state changes affect the coupling relationship between permutation entropy and structural characteristics. The coupling sensitivity calculation formula is as follows: ; : Coupling sensitivity data, : Memory network statistics: variance of sequence sets. Normalization: .

[0069] After obtaining the coupling sensitivity data, it is normalized to ensure that the coupling sensitivity data across different time windows falls within a uniform measurement range. Coupling strength coefficients are then generated based on the normalized coupling sensitivity data. , The value of is limited to the range of 0 to 1. Coupling strength coefficient. Formula generation: ; : No. The coupling strength coefficient of each time window, Normalized coupling sensitivity data .

[0070] In this embodiment, when the network state fluctuates significantly and the entropy-structure correlation is high, Larger values ​​increase the integration ratio of permutation entropy, recurrence rate, and determinism in the calculation of congestion drivers; when the correlation is low, A smaller value ensures that the permutation entropy or structural features remain relatively independent during the fusion process. Congestion driver generation formula: ; Congestion drive quantity Normalized permutation entropy Normalized recurrence rate and certainty : Coupling strength coefficient.

[0071] In generation Subsequently, when the expression ratio control module performs coupling strength adjustment processing, it bases on... The fusion strength of permutation entropy, recurrence rate, and determinism is adjusted to generate a congestion driving force, and a proportional relationship value is generated based on the congestion driving force. The proportional relationship value is used to control the proportional allocation of semantic structure sub-stream data and background texture sub-stream data in the hierarchical compressed image data.

[0072] Subsequently, proportional relationship values ​​in the expression proportional control data are generated based on the congestion driving quantity. These proportional relationship values ​​include the proportional values ​​corresponding to the semantic structure sub-stream data and the corresponding proportional values ​​corresponding to the background texture sub-stream data, and the sum of these two proportional values ​​is 1. The formula for generating the proportional relationship values ​​is: ; ; : The proportional value corresponding to the semantic structure sub-codestream data. : The proportional value corresponding to the background texture sub-stream data. Congestion driver. Satisfies: .

[0073] After generating the expression ratio control data, the expression ratio control module transmits the ratio value to the dual-channel compression generation module. The dual-channel compression generation module adjusts the generation ratio of the semantic structure sub-stream data and the background texture sub-stream data based on the ratio value, ensuring that the ratio of the semantic structure sub-stream data to the background texture sub-stream data in the subsequent layered compressed image data matches the ratio value.

[0074] In this embodiment, after receiving the layered image expression data generated by the layered expression construction module and the expression ratio control data generated by the expression ratio control module, the dual-channel compression generation module performs channel-wise compression and ratio adjustment processing on the layered image expression data to generate semantic structure sub-code stream data, background texture sub-code stream data, and layered compressed image data.

[0075] First, the dual-channel compression generation module extracts semantic structure representation data and background texture representation data from the layered image representation data. Following the acquisition time sequence, the semantic structure representation data and background texture representation data are divided into data blocks, generating sets of semantic structure data blocks and background texture data blocks respectively. The acquisition time sequence is maintained during the data block division process, ensuring that each data block is associated with its corresponding acquisition time information.

[0076] Subsequently, the semantic structure data block set is subjected to independent compression processing to generate semantic structure sub-stream data; the background texture data block set is subjected to independent compression processing to generate background texture sub-stream data. During the generation process, the semantic structure sub-stream data and the background texture sub-stream data maintain the same data block order as the acquisition time order.

[0077] After generating the semantic structure sub-stream data, encryption is performed on the semantic structure sub-stream data to generate encrypted semantic structure sub-stream data. The encryption process only applies to the semantic structure sub-stream data and does not change the data content structure of the background texture sub-stream data.

[0078] After encryption, based on the proportional relationship value in the expression ratio control data, a ratio adjustment process is performed on the encrypted semantic structure sub-stream data and the background texture sub-stream data. This ratio adjustment process controls the allocation ratio of the semantic structure sub-stream data and the background texture sub-stream data in terms of the number of data blocks, data block length, or data block transmission order, ensuring that the proportions of the semantic structure sub-stream data and the background texture sub-stream data in the current generation cycle are consistent with the proportional relationship value. After the ratio adjustment process, adjusted semantic structure sub-stream data and background texture sub-stream data are generated.

[0079] Finally, the adjusted semantic structure sub-stream data and background texture sub-stream data are organized and encapsulated according to the acquisition time sequence. During the encapsulation process, the temporal correspondence between the semantic structure sub-stream data and the background texture sub-stream data is maintained, so that sub-stream data corresponding to the same acquisition time are associated in the encapsulation structure, generating layered compressed image data.

[0080] The generated hierarchical compressed image data is transmitted to the network state sampling module to generate network state sampling data, and simultaneously to the hierarchical reconstruction module to generate semantic structure expression data, background texture expression data, and reconstructed traffic image data.

[0081] In this embodiment, after receiving the layered compressed image data generated by the dual-channel compression generation module, the layered reconstruction module performs sub-code stream parsing and reconstruction processing on the layered compressed image data to generate semantic structure expression data, background texture expression data, and reconstructed traffic image data.

[0082] First, the layered reconstruction module performs sub-stream parsing processing on the layered compressed image data. This parsing process follows the encapsulation structure of the layered compressed image data, separating the semantic structure sub-stream data from the background texture sub-stream data. During parsing, the acquisition time order is maintained, ensuring that the separated semantic structure sub-stream data and background texture sub-stream data maintain the same temporal order as the original encapsulation.

[0083] Subsequently, the semantic structure sub-stream data is decrypted to generate decrypted semantic structure sub-stream data. This decryption process corresponds to the encryption process in the dual-channel compression generation module, restoring the semantic structure sub-stream data to a decompressible data structure.

[0084] After decryption, the decrypted semantic structure sub-stream data is decompressed to generate semantic structure representation data. Simultaneously, the background texture sub-stream data is decompressed to generate background texture representation data. During decompression, the correspondence between data blocks and the acquisition time sequence is maintained, ensuring that the semantic structure representation data and background texture representation data remain consistent with the layered image representation data in the time dimension.

[0085] After obtaining the semantic structure representation data and the background texture representation data, the hierarchical reconstruction module performs spatial alignment processing on the two types of representation data according to the acquisition time sequence of the temporal traffic image data. The spatial alignment processing maps the semantic structure representation data and the background texture representation data to spatial coordinates based on the pixel position relationships corresponding to the structural region division results, ensuring that the structural region positions in the semantic structure representation data are consistent with the corresponding pixel positions in the background texture representation data.

[0086] After spatial alignment, the semantic structure representation data is overlaid onto the corresponding structural regions of the background texture representation data. During the overlay process, the correspondence between region boundary information remains unchanged, ensuring that the structural region locations are provided by the semantic structure representation data and the background region locations are provided by the background texture representation data, thus generating reconstructed traffic image data.

[0087] After generating the reconstructed traffic image data, the hierarchical reconstruction module generates sub-stream reconstruction status information based on the parsing status of the hierarchical compressed image data. This sub-stream reconstruction status information is used to feed back to the network state sampling module, participating in the generation process of the network state sampling data.

[0088] In this embodiment, the network state sampling module records and organizes network transmission and reconstruction behaviors during the transmission and reconstruction of layered compressed image data to generate network state sampling data.

[0089] First, during the transmission of layered compressed image data, the network status sampling module records the transmission status of the layered compressed image data at each acquisition time and generates transmission confirmation information. This confirmation information records whether transmission confirmation has been completed during network transmission of the layered compressed image data and the number of confirmations.

[0090] Subsequently, during the transmission of layered compressed image data, the fragment retransmission behavior corresponding to the layered compressed image data is recorded, generating fragment retransmission information. This fragment retransmission information records the number of fragment retransmissions that occurred during the transmission of the layered compressed image data, as well as the acquisition time information corresponding to each retransmission.

[0091] After the layered reconstruction module completes the parsing and reconstruction of the sub-streams, it generates sub-stream reconstruction status information based on the parsing results of the layered compressed image data. This sub-stream reconstruction status information reflects the reconstruction status of the semantic structure sub-stream data and the background texture sub-stream data during the parsing, decryption, and decompression processes.

[0092] After obtaining the transmission confirmation information, fragment retransmission information, and sub-stream reconstruction status information, the network status sampling module aligns and organizes these three types of information according to the acquisition time sequence of the layered compressed image data. Based on the acquisition time, the transmission confirmation information, fragment retransmission information, and sub-stream reconstruction status information corresponding to the same acquisition time are integrated to form a unified record structure corresponding to the acquisition time.

[0093] After alignment and processing, the network status sampling data is organized into a unified record structure arranged in chronological order of acquisition time. Each record in the network status sampling data contains transmission confirmation information, fragment retransmission information, and sub-stream reconstruction status information corresponding to the acquisition time.

[0094] The generated network state sampling data is transmitted to the expression ratio control module, which generates expression ratio control data based on the network state sampling data. The expression ratio control module then adjusts the ratio of the semantic structure sub-stream data and the background texture sub-stream data in the subsequent layered compressed image data according to the ratio relationship value in the expression ratio control data.

[0095] A method for compression and encryption of traffic images for low-bandwidth network transmission includes the following steps: Acquire continuous traffic image data, and perform time stamping, time continuity verification, and inter-frame consistency processing on the traffic image data according to the acquisition time sequence to generate time-series traffic image data; Semantic target parsing processing is performed on time-series traffic image data to generate semantic category information, structural region division results and region boundary information, and organize them into semantic structure separation data; Based on semantic structure separation data, a set of structural regions is generated. Morphological skeleton extraction is performed on the set of structural regions to generate skeleton topological and geometric descriptions. A skeleton fidelity adjustment coefficient α is introduced to reconstruct the skeleton geometric descriptions with fidelity, generating semantic structure expression data. Simultaneously, a set of background regions is generated based on the set of structural regions. Texture statistical processing is performed on the set of background regions to generate basic background texture expression. Residual extraction is performed on the coverage positions of structural regions to generate structural region residual expression. A texture co-occurrence fusion coefficient β is introduced to perform weighted fusion processing on the basic background texture expression and the structural region residual expression to generate background texture expression data. Finally, the semantic structure expression data and the background texture expression data are aligned according to the acquisition time order to generate layered image expression data. Semantic structure representation data and background texture representation data in the layered image representation data are divided into data blocks and compressed independently to generate semantic structure sub-code stream data and background texture sub-code stream data. The semantic structure sub-code stream data is encrypted and organized into layered compressed image data. During the transmission of layered compressed image data, transmission confirmation information and fragment retransmission information are recorded. The layered reconstruction module generates sub-stream reconstruction status information based on the layered compressed image data. The transmission confirmation information, fragment retransmission information, and sub-stream reconstruction status information are then combined to generate network status sampling data. The network state sampling data is processed by time-series memory aggregation and a time-series memory coefficient γ is introduced to generate a memory network statistical sequence group. The memory network statistical sequence group is then processed by symbolization and permutation pattern statistics to generate the transmission acknowledgment permutation entropy, fragment retransmission permutation entropy, and reconstructed state permutation entropy, and the recurrence rate and determinism are generated. A coupling strength coefficient δ is introduced to couple and adjust the permutation entropy, recurrence rate, and determinism to generate congestion driving force. Based on the congestion-driven quantity, the proportional relationship value in the expression proportional control data is generated, and the proportional relationship between the semantic structure sub-code stream data and the background texture sub-code stream data in the subsequent layered compressed image data is adjusted according to the proportional relationship value. The layered compressed image data is parsed, decrypted, and decompressed to generate semantic structure representation data and background texture representation data. The semantic structure representation data is then overlaid onto the structural regions corresponding to the background texture representation data to generate reconstructed traffic image data.

[0096] Example 1: This example is applied to a video surveillance point at an intersection of a main urban road. The traffic image acquisition module acquires continuous traffic image data at a fixed acquisition frequency, with each frame having a resolution of 1920×1080 pixels. The traffic image acquisition module writes an acquisition time stamp to each frame of traffic image data to generate image unit data; it performs time continuity verification processing on the image unit data, marking and retaining the original data when individual frame acquisition time jumps are detected; then, it rearranges the image unit data according to the acquisition time order to generate time-aligned image data; and it performs inter-frame consistency processing on the time-aligned image data to form time-series traffic image data.

[0097] The semantic structure separation module reads time-series traffic image data frame by frame, extracts pixel data, and establishes frame identifiers. Semantic category determination is performed in each frame, labeling the pixel positions corresponding to vehicles, pedestrians, traffic lights, and road boundaries as different semantic category information. Based on the semantic category information, clustering is performed to form multiple structural regions, generating structural region division results and region boundary information, thus organizing the semantic structure separation data.

[0098] The hierarchical representation construction module generates a set of structural regions based on semantic structure separation data. For vehicle and pedestrian structural regions, morphological skeleton extraction is performed to generate a skeleton pixel set, skeleton topological description, and skeleton geometric description. Based on skeleton boundary deviation data, skeleton structural complexity data, and skeleton coverage ratio data, a skeleton fidelity adjustment coefficient α is generated, and fidelity reconstruction processing is performed on the skeleton geometric description to generate semantic structure representation data. Simultaneously, a background region set is generated, and texture statistical processing is performed on the background region set to generate a basic background texture representation. Residual extraction processing is performed on the structural region coverage positions to generate structural region residual representations. Based on texture difference amplitude data, texture difference distribution density data, and residual embedding ratio data, a texture co-occurrence fusion coefficient β is generated. Weighted fusion is performed on the basic background texture representation and the structural region residual representations to generate background texture representation data. The two types of representation data are aligned according to the acquisition time order to form hierarchical image representation data.

[0099] The dual-channel compression generation module divides the semantic structure representation data and the background texture representation data into data blocks and performs independent compression processing to generate semantic structure sub-codestream data and background texture sub-codestream data respectively; it performs encryption processing on the semantic structure sub-codestream data to generate encrypted semantic structure sub-codestream data; based on the ratio relationship value in the expression ratio control data, it performs ratio adjustment processing on the two types of sub-codestream data to generate layered compressed image data.

[0100] When the wireless link is congested, the network state sampling module records transmission acknowledgment information and fragment retransmission information, and receives sub-stream reconstruction state information generated by the hierarchical reconstruction module, forming network state sampling data. The expression ratio control module performs time-series memory aggregation processing on the network state sampling data, introduces a time-series memory coefficient γ to generate a memory network statistical sequence group; generates transmission acknowledgment permutation entropy, fragment retransmission permutation entropy, reconstruction state permutation entropy, as well as recurrence rate and determinism; introduces a coupling strength coefficient δ to generate congestion driving force; and generates a ratio relationship value based on the congestion driving force, increasing the ratio value corresponding to the semantic structure sub-stream data and decreasing the ratio value corresponding to the background texture sub-stream data, thereby prioritizing the transmission integrity of key semantic structure regions.

[0101] The layered reconstruction module performs sub-stream parsing, decryption, and decompression on the layered compressed image data to generate semantic structure representation data and background texture representation data. The semantic structure representation data is then overlaid onto the corresponding structural regions of the background texture representation data to generate reconstructed traffic image data. This embodiment maintains clear outlines of vehicle and pedestrian structural regions even in low-bandwidth fluctuation environments, while background details dynamically change according to bandwidth.

[0102] Example 2: This example is applied to a remote monitoring node on a highway, connected to the central platform via a dedicated network link. The traffic image acquisition module continuously collects traffic image data and performs time stamping, time continuity verification, and inter-frame consistency processing according to the acquisition time sequence to generate time-series traffic image data.

[0103] The semantic structure separation module performs semantic category determination processing on each frame of traffic image, dividing the pixels corresponding to large vehicles, small vehicles, emergency lane signs and road boundaries into different structural regions, generating structural region division results and region boundary information, forming semantic structure separation data.

[0104] The layered representation construction module performs morphological skeleton extraction processing on the large vehicle structural region, generating skeleton topological and geometric descriptions. Under the current frame identifier, it calculates skeleton boundary deviation data, skeleton structural complexity data, and skeleton coverage ratio data to generate a skeleton fidelity adjustment coefficient α. It then performs fidelity reconstruction processing on the skeleton geometric description to generate semantic structural representation data. Texture statistical processing is performed on the background region set to generate a basic background texture representation. Residual extraction processing is performed on the structural region coverage location to generate a structural region residual representation. Based on texture difference amplitude data, texture difference distribution density data, and residual embedding ratio data, a texture co-occurrence fusion coefficient β is generated. Weighted fusion processing is performed on the two types of representations to generate background texture representation data. Finally, layered image representation data is formed.

[0105] During the stable phase of the private network link, the entropy of the transmission acknowledgment arrangement in the network state sampling data is low, and the reproducibility and determinism are high. The congestion driving quantity generated by the expression ratio control module is at a low level, and the corresponding proportion value of the generated semantic structure sub-code stream data decreases, while the corresponding proportion value of the background texture sub-code stream data increases, thereby improving the expression ratio of texture details in the background region.

[0106] When a sudden fluctuation occurs in the private network link, the network status sampling module records an increase in fragment retransmission information. The expression ratio control module generates a higher congestion driving quantity through time-series memory aggregation processing and coupling strength adjustment processing, dynamically increasing the proportion of semantic structure sub-code stream data, so that key structural areas are transmitted first.

[0107] The layered reconstruction module performs sub-stream parsing, decryption, decompression, and spatial alignment processing on the layered compressed image data to restore semantic structure and background texture representation data, and then overlays this data to generate reconstructed traffic image data. This embodiment ensures the integrity of background details during stable link phases and the stability of vehicle and road structure boundaries during fluctuating link phases, achieving a dynamic balance between transmission quality and semantic fidelity.

[0108] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A low-bandwidth network transmission compression and encryption system for traffic images, characterized in that: include: The traffic image acquisition module is used to acquire continuous traffic image data and organize it into time-series traffic image data according to the acquisition time sequence. The semantic structure separation module performs semantic target parsing processing on time-series traffic image data, divides the time-series traffic image data into structural regions based on semantic category information, and generates semantic structure separation data. The hierarchical representation construction module constructs semantic structure representation data and background texture representation data based on semantic structure separation data, and organizes the semantic structure representation data and background texture representation data into hierarchical image representation data; The expression ratio control module generates expression ratio control data based on network state sampling data, and adjusts the ratio of semantic structure sub-code stream data and background texture sub-code stream data in subsequent layered compressed image data according to the expression ratio control data. The dual-channel compression generation module performs independent compression processing on the layered image representation data to generate semantic structure sub-codestream data and background texture sub-codestream data, and performs encryption processing on the semantic structure sub-codestream data to form layered compressed image data; The layered reconstruction module performs sub-stream parsing on the layered compressed image data, generates sub-stream reconstruction status information, and generates semantic structure expression data and background texture expression data respectively. Based on the semantic structure expression data and background texture expression data, it generates reconstructed traffic image data. The network status sampling module collects transmission confirmation information and fragment retransmission information during the transmission of layered compressed image data, and receives sub-stream reconstruction status information generated by the layered reconstruction module. It then organizes the transmission confirmation information, fragment retransmission information, and sub-stream reconstruction status information to generate network status sampling data.

2. The traffic image low-bandwidth network transmission compression and encryption system according to claim 1, characterized in that: The traffic image acquisition module timestamps the acquired traffic image data according to the acquisition time sequence, generating image unit data with acquisition time information; Perform time continuity verification on the image unit data, mark the image unit data with missing or abnormal jumps in acquisition time, and rearrange the image unit data according to the acquisition time order to generate time-aligned image data; Inter-frame consistency processing is performed on time-aligned image data, traffic image data corresponding to repeated acquisition times are merged, and unique image unit data corresponding to the acquisition time is retained. Image unit data, arranged in chronological order of acquisition and processed for time continuity and inter-frame consistency, are organized into time-series traffic image data.

3. The traffic image low-bandwidth network transmission compression and encryption system according to claim 1, characterized in that: The semantic structure separation module reads the temporal traffic image data frame by frame, extracts the pixel data in each frame, and establishes the corresponding frame identifier. In each frame of the image, semantic category determination processing is performed on the pixel data to generate semantic category information corresponding to the pixel position; Based on semantic category information, pixel locations are clustered and organized to form multiple structural regions. Each structural region is then associated with corresponding semantic category information to generate structural region partitioning results. Perform boundary extraction processing on the structural region division results to generate the region boundary information of each structural region; Semantic category information, structural region division results, and region boundary information are organized by frame identifier to form semantic structure separated data.

4. The traffic image low-bandwidth network transmission compression and encryption system according to claim 1, characterized in that: The hierarchical representation construction module extracts structural region division results, semantic category information, and region boundary information from semantic structure separation data to generate a set of structural regions. The morphological skeleton extraction process is performed on the set of structural regions region by region to generate a skeleton pixel set, and a skeleton topological description and a skeleton geometric description are generated on the skeleton pixel set; Introducing skeleton fidelity adjustment coefficient Based on the skeleton fidelity adjustment coefficient The skeleton geometry description is subjected to fidelity reconstruction processing. The segmented sequence in the skeleton geometry description is simplified or refined to generate a fidelity skeleton geometry description. The region boundary information, skeleton topology description, skeleton fidelity geometry description and semantic category information are organized by region number to form semantic structure expression data. A set of background regions is generated based on a set of structural regions. Texture statistical processing is then performed on the set of background regions to generate a basic representation of the background texture. Perform residual extraction processing on the structural region coverage location to generate structural region residual representation; Introducing texture co-occurrence blending coefficient Based on the texture co-occurrence fusion coefficient The background texture base representation and the structural region residual representation are weighted and fused to generate a fused texture representation, which is then organized into background texture representation data. The weight of the background texture base representation is... The weights of the residual representation in the structural region are ; Semantic structure representation data and background texture representation data are aligned according to the acquisition time sequence of temporal traffic image data, a region number correspondence is established, and layered image representation data is organized.

5. The traffic image low-bandwidth network transmission compression and encryption system according to claim 1, characterized in that: The expression ratio control module organizes the transmission confirmation information, fragment retransmission information, and sub-stream reconstruction status information in the network status sampling data into a network statistical sequence group for the current time window according to the collection time order. Perform temporal memory convergence processing on the current time window network statistical sequence group and the previous time window network statistical sequence group to generate a memory network statistical sequence group, and introduce temporal memory coefficients. The temporal memory aggregation processing is based on the temporal memory coefficient. Adjust the retention rate of the network statistical sequence group in the previous time window; Symbolization and permutation pattern statistical processing are performed on the statistical sequence group of the memory network respectively to generate the permutation entropy of transmission acknowledgment, fragment retransmission, and reconstruction state, and time window normalization is performed on the three types of permutation entropy. A network state vector sequence is constructed based on the statistical sequence group of the memory network, generating the recurrence rate and determinism, and the recurrence rate and determinism are normalized within the time window. The normalized permutation entropy and normalized recurrence rate are subjected to deterministic coupling strength adjustment processing to generate congestion driving forces, and a coupling strength coefficient is introduced. The coupling strength adjustment process is based on the coupling strength coefficient. Adjusting the fusion strength of permutation entropy with recurrence rate and determinism; The proportional relationship value in the expression proportional control data is generated based on the congestion driving quantity. The proportional relationship value includes the proportional value corresponding to the semantic structure sub-code stream data and the proportional value corresponding to the background texture sub-code stream data, and the sum of the two is 1. After generating the expression ratio control data, the generation ratio of the semantic structure sub-codestream data and the background texture sub-codestream data is adjusted according to the ratio relationship value, so that the ratio relationship between the semantic structure sub-codestream data and the background texture sub-codestream data in the subsequent layered compressed image data is consistent with the ratio relationship value.

6. The traffic image low-bandwidth network transmission compression and encryption system according to claim 1, characterized in that: The dual-channel compression generation module divides the semantic structure representation data and background texture representation data in the layered image representation data into data blocks according to the acquisition time order, and generates a set of semantic structure data blocks and a set of background texture data blocks. Perform independent compression processing on the set of semantic structure data blocks to generate semantic structure sub-code stream data; Perform independent compression processing on the background texture data block set to generate background texture sub-stream data; Encryption processing is performed on the semantic structure sub-codestream data to generate encrypted semantic structure sub-codestream data; Based on the proportional relationship value in the expression ratio control data, the ratio adjustment processing of the encrypted semantic structure sub-code stream data and the background texture sub-code stream data is carried out. By controlling the allocation ratio of the semantic structure sub-code stream data and the background texture sub-code stream data in the number of data blocks, the length of data blocks, or the order of data block sending, the adjusted semantic structure sub-code stream data and background texture sub-code stream data are generated. The adjusted semantic structure sub-stream data and background texture sub-stream data are organized and encapsulated according to the acquisition time sequence to generate layered compressed image data.

7. The traffic image low-bandwidth network transmission compression and encryption system according to claim 1, characterized in that: The layered reconstruction module performs sub-stream parsing on the layered compressed image data, separating semantic structure sub-stream data and background texture sub-stream data. Decryption processing is performed on the semantic structure sub-codestream data to generate decrypted semantic structure sub-codestream data; Decompression processing is performed on the decrypted semantic structure sub-code stream data to generate semantic structure representation data; Decompression is performed on the background texture sub-stream data to generate background texture representation data; The semantic structure representation data and background texture representation data are spatially aligned according to the acquisition time sequence of the temporal traffic image data to restore the pixel positions corresponding to the structural region segmentation results. After completing the spatial alignment process, the semantic structure representation data is overlaid onto the structural region corresponding to the background texture representation data to generate reconstructed traffic image data.

8. A traffic image low-bandwidth network transmission compression and encryption system according to claim 1, characterized in that: The network status sampling module records the transmission confirmation information corresponding to the layered compressed image data during the transmission process; Record the fragment retransmission information corresponding to the layered compressed image data during the transmission process; The hierarchical reconstruction module generates sub-stream reconstruction status information based on the hierarchically compressed image data; The confirmation message, fragment retransmission message, and sub-stream reconstruction status information are aligned and organized according to the acquisition time sequence of the layered compressed image data to generate network status sampling data. The network state sampling data is transmitted to the expression ratio control module, which generates expression ratio control data based on the network state sampling data. The expression ratio control module then adjusts the ratio of the semantic structure sub-stream data and the background texture sub-stream data in the subsequent layered compressed image data according to the ratio relationship value in the expression ratio control data.

9. A traffic image low-bandwidth network transmission compression and encryption system according to claim 1, characterized in that: The temporal traffic image data generated by the traffic image acquisition module is transmitted to the semantic structure separation module; The semantic structure separation module generates semantic structure separation data, which is then transmitted to the hierarchical representation construction module. The semantic structure representation data and background texture representation data generated by the hierarchical representation construction module are organized into hierarchical image representation data, and the hierarchical image representation data is transmitted to the representation ratio control module and the dual-channel compression generation module. The network state sampling module generates network state sampling data based on hierarchical compressed image data, and the network state sampling data is transmitted to the expression ratio control module. The expression ratio control module generates expression ratio control data based on hierarchical image expression data and network state sampling data, and the expression ratio control data is transmitted to the dual-channel compression generation module. The dual-channel compression generation module generates semantic structure sub-codestream data and background texture sub-codestream data based on hierarchical image expression data and expression ratio control data, and organizes them into hierarchical compressed image data. The hierarchical compressed image data is then transmitted to the network state sampling module and the hierarchical reconstruction module. The hierarchical reconstruction module generates semantic structure representation data and background texture representation data based on hierarchical compressed image data, and then generates reconstructed traffic image data based on the semantic structure representation data and background texture representation data.

10. A method for applying to the traffic image low-bandwidth network transmission compression and encryption system according to any one of claims 1-9, characterized in that, Includes the following steps: Acquire continuous traffic image data, and perform time stamping, time continuity verification, and inter-frame consistency processing on the traffic image data according to the acquisition time sequence to generate time-series traffic image data; Semantic target parsing processing is performed on time-series traffic image data to generate semantic category information, structural region division results and region boundary information, and organize them into semantic structure separation data; Based on semantic structure separation data, a set of structural regions is generated. Morphological skeleton extraction is performed on the set of structural regions to generate skeleton topological and geometric descriptions. A skeleton fidelity adjustment coefficient α is introduced to reconstruct the skeleton geometric descriptions with fidelity, generating semantic structure expression data. Simultaneously, a set of background regions is generated based on the set of structural regions. Texture statistical processing is performed on the set of background regions to generate basic background texture expression. Residual extraction is performed on the coverage positions of structural regions to generate structural region residual expression. A texture co-occurrence fusion coefficient β is introduced to perform weighted fusion processing on the basic background texture expression and the structural region residual expression to generate background texture expression data. Finally, the semantic structure expression data and the background texture expression data are aligned according to the acquisition time order to generate layered image expression data. Semantic structure representation data and background texture representation data in the layered image representation data are divided into data blocks and compressed independently to generate semantic structure sub-code stream data and background texture sub-code stream data. The semantic structure sub-code stream data is encrypted and organized into layered compressed image data. During the transmission of layered compressed image data, transmission confirmation information and fragment retransmission information are recorded. The layered reconstruction module generates sub-stream reconstruction status information based on the layered compressed image data. The transmission confirmation information, fragment retransmission information, and sub-stream reconstruction status information are then combined to generate network status sampling data. The network state sampling data is processed by time-series memory aggregation and a time-series memory coefficient γ is introduced to generate a memory network statistical sequence group. The memory network statistical sequence group is then processed by symbolization and permutation pattern statistics to generate the transmission acknowledgment permutation entropy, fragment retransmission permutation entropy, and reconstructed state permutation entropy, and the recurrence rate and determinism are generated. A coupling strength coefficient δ is introduced to couple and adjust the permutation entropy, recurrence rate, and determinism to generate congestion driving force. Based on the congestion-driven quantity, the proportional relationship value in the expression proportional control data is generated, and the proportional relationship between the semantic structure sub-code stream data and the background texture sub-code stream data in the subsequent layered compressed image data is adjusted according to the proportional relationship value. The layered compressed image data is parsed, decrypted, and decompressed to generate semantic structure representation data and background texture representation data. The semantic structure representation data is then overlaid onto the structural regions corresponding to the background texture representation data to generate reconstructed traffic image data.

Images