Image efficient compression and rapid transmission method and device in vehicle damage claim settlement

Abstract

The invention relates to the technical field of image processing, and provides an image efficient compression and rapid transmission method and device in vehicle damage claim settlement, and the method comprises the steps: obtaining a vehicle damage site original image through a mobile terminal device with an image collection function; identifying and positioning a vehicle damage key area in the original image of the vehicle damage site by adopting a multi-mode identification mode to obtain position coordinates and area range information of the vehicle damage key area; based on the position coordinates and the area range information of the vehicle damage key area, carrying out hierarchical compression processing on the vehicle damage field image; according to a network environment monitored in real time, carrying out adaptive quantization on the image data after hierarchical compression; the quantized image data is processed based on a preset transmission optimization protocol and then uploaded to a server, so that the server decodes and restores the image data; and receiving the decoded and restored standard format image returned by the server, thereby realizing high-fidelity compression and rapid transmission of the vehicle damage image.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of image processing, in particular to an image efficient compression and fast transmission method and device in vehicle damage claim settlement. BACKGROUND

[0002] With the acceleration of the digitalization process of vehicle insurance claim settlement business, efficient processing of vehicle damage scene images has become a key link restricting the efficiency of claim settlement. In the traditional vehicle damage claim settlement scenario, users need to upload vehicle damage photos to the server through mobile terminals. However, due to the large data volume of a single high-resolution image (such as 4K / 8K) when uncompressed, the following problems may occur: Storage space pressure: direct storage of uncompressed images will occupy a large amount of server resources. According to the calculation of daily processing of ten thousand cases, the annual storage cost may exceed one million yuan; Transmission delay: in a mobile network environment, the transmission time of large-size images may take tens of seconds, affecting user experience; Waste of computing power: the server needs additional resources for image preprocessing, reducing the overall claim settlement system throughput.

[0003] Although existing technologies (such as JPEG compression) can reduce file size, they have the problem of detail loss caused by lossy compression, and formats such as WebP have insufficient compatibility, making it difficult to meet the comprehensive needs of vehicle insurance scenarios for image accuracy and transmission efficiency. SUMMARY

[0004] The present disclosure provides an image efficient compression and fast transmission method and device in vehicle damage claim settlement, aiming to realize high-fidelity compression and fast transmission of vehicle damage images.

[0005] According to a first aspect of the present disclosure, an image efficient compression and fast transmission method in vehicle damage claim settlement is provided, comprising: acquiring a vehicle damage scene original image through a mobile terminal device with image acquisition function; identifying and positioning a vehicle damage key area in the vehicle damage scene original image by using a multi-modal recognition method to obtain position coordinates and area range information of the vehicle damage key area; performing layered compression processing on the vehicle damage scene image based on the position coordinates and area range information of the vehicle damage key area; performing adaptive quantization on the layered compressed image data according to real-time monitoring of the network environment; uploading the quantized image data to the server after processing based on a preset transmission optimization protocol, so that the server decodes and restores the image data; receiving the standard format image returned by the server after decoding and restoring.

[0006] As a preferred embodiment, a multi-modal recognition method is used to recognize and locate the vehicle damage key region in the vehicle damage scene original image, to obtain the position coordinates and region range information of the vehicle damage key region, including: The vehicle damage scene original image is input into a pre-trained deep target detection model, local features are extracted by CNN, multi-scale features are fused by FPN, and preliminary position coordinates of the vehicle damage key region are output; The vehicle body contour is located by an edge detection algorithm, the edge features of the vehicle damage key region are enhanced by morphological dilation and erosion operations, and the collision points are located by Hough transform; The preliminary position coordinates of the vehicle damage key region, the edge features and the collision point information are fused to output the accurate position coordinates and region range parameters of the vehicle damage key region.

[0007] As a preferred embodiment, the vehicle damage scene image is compressed in layers based on the position coordinates and region range information of the vehicle damage key region, including: The vehicle damage scene image is decomposed in multiple scales using a biorthogonal wavelet basis to generate a low-frequency approximation component and multiple high-frequency detail components, wherein the high-frequency detail components include high-frequency detail components corresponding to the vehicle damage key region and high-frequency detail components corresponding to non-vehicle damage key regions; Based on the position coordinates and region range information of the vehicle damage key region, the low-frequency approximation component and the multiple high-frequency detail components are compressed in layers respectively.

[0008] As a preferred embodiment, the low-frequency approximation component and the high-frequency detail component are compressed in layers based on the position coordinates and region range information of the vehicle damage key region, including: The low-frequency approximation component is compressed using a lossless compression strategy to preserve the basic information of the image; The high-frequency detail components corresponding to the vehicle damage key region are retained in their entirety by the position coordinates and region range information of the vehicle damage key region and are assigned independent code streams; The significant coefficients of the high-frequency detail components corresponding to the non-vehicle damage key regions are selectively truncated.

[0009] As a preferred embodiment, the image data compressed in layers is adaptively quantized according to the real-time monitored network environment, including: The network environment parameters are obtained in real time through a mobile terminal network interface, wherein the network environment parameters at least include network type, real-time bandwidth and delay jitter; According to the real-time obtained network environment parameters and the pre-set mapping relationship between network quality and quantization step length, the corresponding quantization step length is automatically switched; When the network type switching or bandwidth fluctuation exceeds the preset fluctuation threshold or the delay jitter exceeds the preset time threshold, a gradual adjustment strategy is adopted to dynamically adjust the quantization step.

[0010] As a preferred embodiment, the quantized image data is uploaded to the server after being processed based on a preset transmission optimization protocol, including: Intelligently fragmenting the quantized image data based on the maximum transmission unit of the network; Dividing the fragmented image data into three priority data packets, wherein the high-frequency detail component corresponding to the vehicle damage key area is a high-priority data packet, the low-frequency approximation component is a medium-priority data packet, and the high-frequency detail component corresponding to the non-vehicle damage key area is a low-priority data packet; According to the pre-set priority scheduling strategy, the data is transmitted to the server through the three-level priority queue set by the transmission layer.

[0011] As a preferred embodiment, according to the pre-set priority scheduling strategy, the data is transmitted to the server through the three-level priority queue of the transmission layer, including: According to the pre-set priority scheduling strategy, the high-priority data packet is inserted into the head of the transmission queue, the medium-priority data packet is inserted into the middle of the queue according to the arrival order, and the low-priority data packet is inserted into the tail of the transmission queue. Using the BBR algorithm to dynamically adjust the sending rate, when network congestion occurs, the high-priority data packet still maintains a minimum of 50% bandwidth reservation.

[0012] As a preferred embodiment, the method further comprises: Forward error correction encoding is performed on the high-priority data packet to generate a preset proportion of redundant data packets, wherein the redundant data packets inherit the priority mark of their corresponding original data packets. When the network packet loss rate exceeds the preset threshold, the redundant data packet priority transmission mechanism is triggered.

[0013] As a preferred embodiment, the server decodes and restores the image data, including: Using context-based adaptive binary arithmetic coding to decode the image data; Implementing wavelet reconstruction on the decoded image data, and performing inverse transformation through a prediction step and an update step; According to the PNG standard, the inverse transformed image data is packaged into a standard format. According to a second aspect of the present disclosure, an image efficient compression and rapid transmission device in vehicle damage claim is provided, comprising: An image acquisition module for acquiring original images of vehicle damage scenes through a mobile terminal device with image acquisition function; The vehicle damage area identification and positioning module is configured to identify and position a vehicle damage key area in the vehicle damage scene original image by using a multi-modal identification method, to obtain position coordinates and area range information of the vehicle damage key area. The image hierarchical compression module is configured to perform hierarchical compression processing on the vehicle damage scene image based on the position coordinates and area range information of the vehicle damage key area. The image adaptive quantization module is configured to perform adaptive quantization on the hierarchical compressed image data according to a real-time monitored network environment. The image data transmission module is configured to upload the quantized image data to a server after processing based on a preset transmission optimization protocol, so that the server decodes and restores the image data. The standard format image generation module is configured to receive the decoded and restored standard format image returned by the server.

[0014] According to a third aspect of the present disclosure, an electronic device, at least one processor, and a memory connected in communication with the at least one processor are provided; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the method of any one of the above.

[0015] Compared with the prior art, the present disclosure has the following beneficial effects: (1) The storage cost is significantly reduced: the present disclosure uses a biorthogonal wavelet basis to perform 3-level decomposition, losslessly compresses low-frequency approximation components, differentially processes high-frequency components according to vehicle damage key areas (ROIs) (more significant coefficients are retained in key areas), and optimizes the quantization step length through a non-uniform quantization table. After compression, the image volume is reduced to 5%-10% of the original image, and the annual storage cost is reduced by more than 80%.

[0016] (2) Transmission efficiency and reliability are improved in weak network environment: the present disclosure dynamically adjusts the quantization step length by real-time monitoring of network type, network bandwidth, and delay jitter, and avoids image quality mutation through a smooth transition mechanism, thereby improving transmission speed and quality, shortening the transmission time to within 300ms in weak network environment, and reducing user waiting time by 60%; (3) Intelligent identification of vehicle damage key areas: the present disclosure is based on the YOLOv10 algorithm combined with traditional image processing (Canny edge detection, Hough transformation), and is trained on 11,675 labeled images to accurately position the collision point, with a collision point positioning error ≤2cm, a damage type classification accuracy ≥92%, a false detection rate ≤5%, and a missed detection rate ≤3%.

[0017] (4) Image quality assurance: ROI area protection mechanism, high-frequency components of key areas of vehicle damage (such as collision points) are allocated independent code streams and more significant coefficients are reserved, while high-frequency components of non-key areas are selectively truncated according to rules, and a non-uniform quantization table is constructed, the quantization step of the edge gradient area (detected by Sobel operator) is reduced by 40%, ensuring that the vehicle damage details are clear and identifiable, and finally the vehicle damage detail SSIM index is ≥0.95, meeting the accuracy requirements of damage determination for vehicle insurance claims.

[0018] (5) Compatibility optimization: The final output of the present disclosure is a standard PNG format image, which can directly interface with the image analysis interface of the existing claims system without additional development of adaptation modules; at the same time, based on NPU parallel computing (wavelet transform speed is improved by 3 times) and asynchronous task scheduling (UI stuttering rate is reduced by 90%), 1080P image full-process processing is ≤150ms, 4K image is ≤300ms on Snapdragon 7 Gen3 and other mid-end models, compatible with more than 95% of mainstream mobile devices.

[0019] (6) Network adaptability enhancement: The present disclosure uses dynamic quantization step and smooth transition mechanism to ensure that there is no sudden change in image quality when the network type is switched or the bandwidth fluctuates; through three priority transmission queues combined with redundant coding, vehicle damage details can still be completely restored under 5% packet loss rate, and can adapt to complex network environments such as urban, suburban, and underground garage.

[0020] It should be understood that the content described in the summary section is not intended to limit the key or important features of the embodiments of the present disclosure, nor to limit the scope of the present disclosure. Other features of the present disclosure will become apparent through the following description. BRIEF DESCRIPTION OF DRAWINGS

[0021] The above and other features, advantages, and aspects of embodiments of the present disclosure will become more apparent by describing in detail the following embodiments with reference to the attached drawings. The drawings are intended to better understand the present disclosure and do not limit the present disclosure. In the drawings, the same or similar reference numerals refer to the same or similar elements, and Figure 1 A flowchart of an image efficient compression and fast transmission method in vehicle damage claims of an embodiment of the present disclosure is shown; Figure 2 A schematic diagram of an image efficient compression and fast transmission system in vehicle damage claims of an embodiment of the present disclosure is shown; Figure 3 A block diagram of an image efficient compression and fast transmission device in vehicle damage claims of an embodiment of the present disclosure is shown; Figure 4 A schematic diagram of an exemplary electronic device of an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0022] The technical solutions and advantages of the embodiments of the present disclosure will be described clearly and completely below with reference to the drawings of the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present disclosure.

[0023] In addition, the term "and / or" herein is only used to describe the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent three cases of A alone, A and B together, and B alone. In addition, the character " / " herein generally represents an "or" relationship between the front and rear associated objects.

[0024] As Figure 1 FIG. 1 is a flowchart of a method for efficient image compression and fast transmission in vehicle damage claim according to an embodiment of the present disclosure. The method 100 comprises the following steps: S110: obtaining vehicle damage scene original images by a mobile terminal device with image acquisition function.

[0025] In some embodiments, the mobile terminal device with image acquisition function described above can be a mobile phone or a tablet device, etc. When collecting images on the scene, after opening the vehicle damage claim special applet or APP on the mobile terminal device, the rear camera is automatically called to enter the collection mode. The APP interface displays the shooting guide frame in real time. The user aligns the damaged part according to the voice prompt (such as "please shoot the front face damage area of the vehicle"). The device automatically adjusts the exposure parameters through the light sensor, and reduces the influence of handheld jitter by using optical anti-shake function. When shooting, at least 3 images of different angles such as front view, side view, and detail close-up view should be continuously shot for single damage. The collected vehicle damage scene original images are temporarily stored in the local cache directory of the device in JPEG format. Each image automatically adds shooting time, GPS position and other metadata.

[0026] S120: identifying and positioning the vehicle damage key area in the vehicle damage scene original image by using a multi-modal recognition method, to obtain the position coordinates and area range information of the vehicle damage key area.

[0027] In some embodiments, the multi-modal recognition method is used to identify and position the vehicle damage key area in the vehicle damage scene original image, which specifically includes the following steps: (a) preliminary detection by deep learning: First, the original image of the vehicle damage scene is input into the pre-trained YOLOv10 deep target detection model. The local texture features of the image (such as damage edges and color mutation areas) are extracted by CNN, and the feature maps of 4 scales (80x80, 40x40, 20x20, 10x10) are fused by FPN (Feature Pyramid Network), and the preliminary position coordinates (x1, y1, x2, y2) of the vehicle damage key area are output.

[0028] (b) Traditional visual feature enhancement and positioning: First, the vehicle body contour is located by the adaptive threshold Canny algorithm, and the noise edges with a gray change less than 20 are filtered out. Second, morphological operations are performed on the edge image: first, a 3x3 rectangular convolution kernel is used for 1 time dilation operation to connect the broken edges, and then a convolution kernel of the same size is used for 1 time erosion operation to remove burrs and enhance the continuous edge features of the vehicle damage area. Finally, the probability Hough transform is used to detect the straight lines of the vehicle body, and the collision point coordinates are located through the line slope mutation point (slope change amount ≥ 30°) and the line segment interruption position.

[0029] (c) Multi-source information fusion and accurate output Finally, the preliminary position coordinates of the vehicle damage key area, the edge features, and the collision point coordinates are fused to output the accurate position coordinates (pixel-level range) and the area range parameters (width, height) of the vehicle damage key area, with a false detection rate ≤5% and a missed detection rate ≤3%.

[0030] In addition, the training and evaluation details of the YOLOv10 deep target detection model include: Training data set: contains more than 10000 labeled images, covering 12 common vehicle types (sedan / SUV / truck, etc.), 8 damage types (such as scratch 32%, dent 28%, crack 15%, deformation 12%, paint loss 8%, glass breakage 3%, lamp damage 1.5%, and other 0.5%), and containing 30% complex scene samples (such as rain reflection, backlight shooting, small damage <5mm, and multiple damage overlapping areas); Training strategy: cosine annealing learning rate scheduling (set initial lr=0.01, iterate 300 times) is adopted, combined with Mosaic data enhancement to strengthen random transformation, and distributed GPU cluster training optimization is adopted; model lightweight technology is adopted to compress the weight file to adapt to mobile deployment.

[0031] Evaluation index: The recognition performance is quantified by mAP (average precision mean) and IoU (intersection over union). On the independent test set, the key indicators (mAP@0.5, mAP@0.5:0.95) reach more than 94% and more than 82% respectively, the false detection rate and the missed detection rate are controlled at a low level (≤5%, ≤3%), and the single image inference time meets the real-time demand of mobile terminal (≤80ms).

[0032] By the above multi-modal recognition method, the key area of vehicle damage in the original image of the vehicle damage scene is recognized and positioned, the preliminary position coordinates, edge features and collision point coordinates of the key area of vehicle damage are fused, and the accurate position coordinates and area range of the key area of vehicle damage are output, which not only plays the strong recognition ability of deep learning on complex damage, but also guarantees the geometric positioning accuracy by using traditional visual algorithm, and realizes efficient and accurate positioning of the key area of vehicle damage.

[0033] S130: Based on the position coordinates and area range information of the key area of vehicle damage, the vehicle damage scene image is processed by hierarchical compression.

[0034] In some embodiments, the hierarchical compression processing of the vehicle damage scene image is realized by spatial domain and frequency domain cooperative optimization, specifically including: (a) Spatial domain layering and wavelet decomposition A biorthogonal wavelet basis with symmetry and compact support is used to perform 3-level multi-scale decomposition on the vehicle damage scene image, generating one low-frequency approximation component (LL3) and six high-frequency detail components (HLi, LHi, HHi, i=1, 2, 3). Among them, the high-frequency detail components are further divided into high-frequency detail components corresponding to the key area of vehicle damage (including core features such as collision points and damage edges) and high-frequency detail components corresponding to non-key areas of vehicle damage (such as smooth areas without damage on the vehicle body). This decomposition method balances the time-frequency resolution, reduces image boundary distortion, and lays the foundation for subsequent differential compression.

[0035] Further, on this basis, the texture layer and the semantic layer are divided: Texture layer: Extract the LL3 low-frequency component, which accounts for more than 85% of the energy of the original image, as the carrier of the image basic outline and overall texture, and use the 5 / 3 lifting wavelet algorithm for lossless compression to ensure that the image macro features are not lost; Semantic layer: Taking the high-frequency detail component as the core, combined with the collision point area coordinates positioned by the YOLOv10 detection model, the high-frequency component corresponding to the key area of vehicle damage is separately labeled and assigned an independent code stream to realize accurate preservation of damage details.

[0036] (b) Differential compression strategy Different features of the texture layer and the semantic layer are processed differently: Texture layer processing: On the basis of lossless compression, a fine quantization strategy of AQ=5 is adopted for the LL3 low-frequency component to retain more than 95% of the energy, control the growth of data volume, and ensure the integrity of the basic information such as the overall contour and color tone of the vehicle body; Semantic layer processing: Based on the position and range information of the vehicle damage key area, the high-frequency detail components corresponding to the vehicle damage key area are completely retained (to avoid loss of details); The high-frequency detail components corresponding to the non-vehicle damage key area are selectively truncated according to the level, for example, the HHi component (containing high-frequency vertical details) retains the top 20% significant coefficients, the Hli / LHi component (containing high-frequency horizontal / diagonal details) retains the top 30% significant coefficients, and the rest of the coefficients are set to zero, which maximizes the compression efficiency under the premise of ensuring the visual acceptability of non-key areas.

[0037] (c) Frequency domain quantization optimization To further balance detail retention and compression ratio, a non-uniform quantization table is constructed: the edge gradient area is detected by the Sobel operator, and the quantization step is reduced by 40% to enhance detail retention and improve the clarity of fine textures; the quantization step of the smooth area without damage (such as the complete surface of the vehicle paint) is increased by 60%, and the compression efficiency is improved by reducing redundant information.

[0038] Through the above-mentioned collaborative strategy of spatial domain layering, differential compression, and frequency domain quantization, the image data can be compressed to 5%-10% of the original size while retaining the detail features of the vehicle damage key area (SSIM index ≥ 0.95), which meets the needs of mobile transmission and provides high-fidelity image basis for subsequent claim settlement and damage assessment.

[0039] S140: Adaptive quantization of the layered compressed image data according to the real-time monitored network environment.

[0040] In some embodiments, the layered compressed image data is adaptively quantized through a three-level mechanism of "network state awareness-quantization parameter dynamic matching-image feature weighted optimization", and the specific process is as follows: (a) Network state awareness Real-time bandwidth and type detection: Through the mobile terminal system interface (such as ConnectivityManager of Android system and NEHotspotNetwork of iOS), the network type (5G / 4G / Wi-Fi / weak network) is obtained in real time, and the actual available bandwidth (Mbps) is calculated by the sliding window algorithm; Delay Jitter Monitoring: Periodic ping to server using ICMP protocol to measure Round Trip Time (RTT), and calculate Jitter by standard deviation of 10 consecutive RTT. When Jitter > 50ms, the network is considered unstable. Network Quality Grading: Based on three-dimensional parameters of bandwidth (high / medium / low), delay (<100ms / 100-300ms / >300ms), and jitter (stable / fluctuating / severe fluctuation), the network state is divided into three levels: good, medium, and poor, providing a basis for quantitative adjustment.

[0041] (b) Dynamic Quantization Step Adjustment Strategy Basic Threshold Mapping: Preset mapping relationship between network level and quantization step (Q value): 5G / Wi-Fi (good) corresponds to Q=10 (fine-grained quantization), 4G (medium) corresponds to Q=30 (balanced quantization), and weak network / high jitter (poor) corresponds to Q=50 (high compression quantization), ensuring balanced transmission efficiency and image quality under different networks. Smooth Transition Mechanism: When the network state changes (such as 4G to Wi-Fi or signal mutation), an exponential decay adjustment strategy is used, with a change amplitude of ≤5 per second (e.g., from Q=50→Q=10 requires 8 seconds of gradual change), avoiding image block effects or fuzzy jumps caused by parameter mutations. Dynamic Threshold Correction: Combined with the current data volume to be transmitted (such as the total size of vehicle damage images), the mapping table is corrected in real time-when the data volume > current bandwidth ×3s transmission volume, automatically increase Q value by 10% to prioritize transmission completion.

[0042] (c) Quantization Matrix Optimization for Vehicle Damage Feature Perception Key Area Weighted Quantization: Based on the key area coordinates of vehicle damage located in S120, a customized non-uniform quantization matrix is used for high-frequency components (such as collision edges and scratch textures) covered by the key area: the quantization step of edge gradient area (detected by Sobel operator) is reduced by 40% (ΔQ=5), ensuring clear damage details; the quantization step of non-key area smooth area (such as complete paint surface) is increased by 60% (ΔQ=15), maximizing redundant compression; Texture Complexity Adaptation: Calculate local texture entropy (Entropy) of the image through Gray Level Co-occurrence Matrix (GLCM), and automatically reduce the quantization step by 20% for complex areas with entropy value >0.8 (such as multiple damage overlapping areas), and increase by 30% for smooth areas with entropy value <0.3, achieving precise control of "complex areas preserving details, simple areas improving efficiency".

[0043] Through the above mechanism, the end-to-end transmission delay can be controlled within 500 ms in a 4G environment, the compression ratio is improved by 30% in a weak network environment, and the SSIM index of the vehicle damage key area remains ≥0.92, balancing the transmission real-time and the detail integrity required for damage determination.

[0044] S150: upload the quantized image data to the server after processing based on the preset transmission optimization protocol, so that the server decodes and restores the image data.

[0045] In some embodiments, the transmission optimization and decoding restoration process is realized through the cooperation of "intelligent transmission layer optimization-server-side accurate reconstruction-mobile terminal hardware acceleration", which specifically includes the following steps: First, based on the network maximum transmission unit, the quantized image data is intelligently fragmented; Second, the fragmented image data is divided into three priority data packets, wherein the high-frequency detail components corresponding to the vehicle damage key area are high-priority data packets, the low-frequency approximate components are medium-priority data packets, and the high-frequency detail components corresponding to the non-vehicle damage key area are low-priority data packets; Finally, according to the pre-set priority scheduling strategy, the three-priority queue set by the transmission layer is used to schedule and transmit data to the server.

[0046] Specifically, (a) intelligent fragmentation strategy Dynamic fragmentation size: adaptively adjust the fragmentation size (usually 1024-1460 bytes) according to the network MTU (maximum transmission unit) to avoid IP layer fragmentation; Last piece filling mechanism: the last data packet that is not enough for fragmentation size is filled with zeros to the standard fragmentation size, and a "last piece identification bit" is added to the packet header to ensure that the server can accurately identify the effective data length and avoid redundant filling interference with decoding.

[0047] Fragmentation verification mechanism: each fragment carries a 16-bit fragment sequence number and a 32-bit overall data hash value, and the server verifies the data integrity through sequence number sorting and hash verification.

[0048] (b) three-priority scheduling strategy According to the pre-set priority scheduling strategy, use the "header insertion-middle ordered-tail cache" strategy to insert high-priority data packets into the head of the transmission queue for priority transmission, insert medium-priority packet data packets into the middle of the queue according to the generation order, and insert low-priority data packets into the tail of the transmission queue (transmitted when the network is idle); Integrated BBR (Bottleneck Bandwidth and RTT) congestion algorithm, real-time detection of network bottleneck bandwidth and round-trip delay, normal network, according to the bandwidth ratio of "high: medium: low = 5:3:2"; when the network is congested (RTT> 300ms), trigger priority protection mechanism: high priority packet reserves the lowest 50% bandwidth, medium priority reduces to 30%, and low priority suspends sending and caches locally (follow-up transmission after network recovery).

[0049] It should be emphasized that in this embodiment, the forward error correction (FEC) enhancement mechanism is also cited, which encodes the high-priority data packets with forward error correction, generates a predetermined proportion of redundant data packets, and specifically, The high-priority data packets are encoded with RS (20, 16), generating a predetermined proportion of redundant packets, for example: 4 redundant packets are generated for every 16 original packets, with a redundancy rate of 20%, and the generated redundant data packets inherit the priority markers of their corresponding original data packets, tolerating up to 4 packet losses in a single group; no redundancy is added to medium / low priority data packets to balance transmission efficiency; Then, through RTT fluctuation and data packet ACK feedback, the packet loss rate is calculated in real time: when the packet loss rate is ≤5%, only the original packet is transmitted; when the packet loss rate is >5%, the redundant packet is automatically triggered for priority transmission, and the server side recovers the lost data in real time through the redundant packet to avoid retransmission delay; Finally, data integrity verification is performed, and a 32-bit CRC check code is integrated into the header of each data packet. The server side verifies the check code first after receiving it, and if it fails, it is marked as "invalid packet" and decides whether to request retransmission according to the priority (high-priority packets must be retransmitted, and low-priority packets can be ignored).

[0050] In some embodiments, the server decoding and restoring the image data includes the following steps: The image data is decoded using context-based adaptive binary arithmetic coding; the decoded image data is subjected to wavelet reconstruction, and inverse transformation is performed through a prediction step and an update step; the inverse transformed image data is encapsulated into a standard format according to the PNG standard, and the specific process is as follows: (a) The image data is decoded using context-based adaptive binary arithmetic coding, specifically including: (a1) Fragment recombination and integrity verification After receiving the fragment data packet uploaded by the mobile terminal, first, according to the fragment sequence number (16 bits), the ordered recombination is performed, and the integrity is verified in combination with the overall data hash value (32 bits). If there is a high-priority fragment loss (such as vehicle damage critical area data), it is recovered immediately through the RS redundant packet (20% redundancy rate); when the redundant packet is insufficient, a directional retransmission request is triggered to ensure the completeness of the bitstream.

[0051] (a2) Context modeling and probability optimization For the feature differences of car damage images (such as high-frequency coefficients of damage edges and low-frequency coefficients of smooth areas of car bodies), 128 special context models are preset: 32 high-frequency coefficient models are assigned to edge gradient areas (Sobel operator markers) to adapt to the distribution of dramatic changes in signs; 16 low-frequency coefficient models are assigned to smooth areas in non-critical areas to optimize the coding efficiency of continuous zero-value signs.

[0052] (a3) Hardware-accelerated interval division GPU parallel computing is used to accelerate the interval division process: the bit stream is split into 64-byte blocks, and interval sub-division is performed in parallel by CUDA cores based on sign probability. Compared with pure CPU decoding, the decoding speed is improved by 3 times, and the decoding time of a single 1080P image is ≤80ms.

[0053] (b) Wavelet reconstruction is performed on the decoded image data, and inverse transformation is performed through prediction and update steps, including: Prediction step: predict even samples from odd samples: d[n] = s[2n+1] - (s[2n] + s[2n+2]) / 2; Update step: correct the approximation coefficient with the detail coefficient: s[n] = s[n] + (d[n-1] + d[n]) / 4; (b1) Cascade processing: inverse transform from the 3rd high-frequency component, gradually merge into LL3 component, and finally reconstruct the original resolution image; (b2) Boundary distortion suppression: use "mirror extension + smooth transition" strategy for pixels outside the image boundary - fill 1-3 pixels outside the boundary with mirror pixels, and process 4-6 pixels with linear decay to avoid edge artifacts caused by traditional periodic extension.

[0054] (c) Encapsulate the inverse-transformed image data into standard PNG format according to the PNG standard, including: (c1) PNG efficient encapsulation: build file structure according to PNG specification: Header: write 8-byte standard signature (\x89PNG\x0D\x0A\x1A\x0A); Data block: add IHDR block (including width, height, bit depth 24bit, color type RGB), IDAT block (compressed image data), and IEND block (end identifier) in sequence; Compression optimization: DEFLATE algorithm (compression level 6) is adopted for IDAT data, combined with the local correlation of vehicle damage images (such as large areas of vehicle paint), the compression ratio is improved by 15% compared with general images; (c2) Metadata embedding: embedding compression parameters (such as quantization step size, wavelet basis type) in tEXt auxiliary blocks to support subsequent damage assessment systems to trace the image processing process.

[0055] In addition, it is also emphasized that in this embodiment, the processing efficiency is also improved through mobile terminal hardware acceleration optimization, which includes the following two aspects: (1) NPU parallel computing acceleration Operator depth adaptation: mapping wavelet transform, quantization and other computationally intensive operations to Snapdragon 7 Gen3 Hexagon NPU: Convolution kernel parallelism: using the 128-bit vector operation capability of HVX (Hexagon Vector eXtensions) unit, the wavelet basis convolution operation is decomposed into 8-way parallel computation, and the wavelet transform time of a single 1080P image is reduced from 200ms to 65ms; Memory efficient reuse: through the NPU shared virtual memory (SVM) mechanism, direct access to the image buffer area collected by the camera (avoiding CPU-NPU data copying), memory access delay is reduced by 70%; Energy efficiency ratio optimization: dynamically adjusting the NPU working frequency (400MHz when idle, 800MHz when fully loaded), the average power consumption of 1080P image compression is 280mW, which is reduced by 55% compared with pure CPU processing.

[0056] (2) Asynchronous task scheduling mechanism Adopting a thread architecture that separates the three threads of "collection-compression-transmission", real-time performance and user experience are ensured through fine scheduling, and the specific design is as follows: Thread decoupling design: adopting a "collection-compression-transmission" three-thread separation architecture: Producer thread (image collection): get a frame of image every 33ms (30fps), store it in a blocking queue with a capacity of 4 (to avoid buffer overflow); Consumer thread (compression processing): after taking the frame from the queue, call the NPU accelerated compression interface, and after processing, send it to the transmission thread through Android Handler / iOS GCD; Transmission thread: asynchronously upload data according to the priority queue, without blocking the compression process; Priority protection: Through thread priority inversion detection (when a low-priority compression thread blocks a high-priority UI thread), the compression thread priority is automatically elevated to "background highest", ensuring that the response time of user interactions such as clicking and zooming is ≤100ms.

[0057] Through the synergistic optimization of NPU hardware acceleration and asynchronous scheduling, the "collection-compression-transmission" whole process delay of a single frame of 1080P image on a mobile terminal is controlled within 150ms, and the delay of 4K image is ≤300ms, while the power consumption is reduced by more than 50%, meeting the dual needs of real-time and device endurance in the vehicle damage claim scenario.

[0058] Through the synergistic design of the above transmission optimization, decoding reconstruction, and hardware acceleration, the transmission success rate of quantized data in a weak network environment is improved to 98%, the SSIM error of the image decoded and restored on the server side and the original image is ≤3%, and the damage size measurement deviation is ≤2mm, fully meeting the professional requirements of vehicle insurance damage for detail clarity, and the single frame processing delay on the mobile terminal is controlled within 150ms, meeting the real-time and accuracy requirements of the vehicle insurance damage scenario.

[0059] S160: Receive the decoded and restored standard format image returned by the server.

[0060] According to the above embodiments of the present disclosure, the following technical effects are achieved: (1) The storage cost is significantly reduced: the present disclosure uses a biorthogonal wavelet basis for 3-level decomposition, lossless compression of low-frequency approximation components, and differential processing of high-frequency components according to vehicle damage key regions (ROI) (more significant coefficients are retained in key regions), and the quantization step is optimized through a non-uniform quantization table, the compressed image volume is reduced to 5%-10% of the original image, and the annual storage cost is reduced by more than 80%.

[0061] (2) Transmission efficiency and reliability are improved in weak network environment: the present disclosure dynamically adjusts the quantization step by monitoring the network type, network bandwidth, and delay jitter in real time, and avoids image quality mutation through a smooth transition mechanism, improving transmission speed and quality, achieving transmission time within 300ms in weak network environment, and reducing user waiting time by 60%; (3) Intelligent identification of vehicle damage key regions: the present disclosure is based on YOLOv10 algorithm combined with traditional image processing (Canny edge detection, Hough transform), trained on 11,675 labeled images, accurately positioning the collision point, achieving collision point positioning error ≤2cm, damage type classification accuracy ≥92%, false positive rate ≤5%, and false negative rate ≤3%.

[0062] (4) Image quality assurance: ROI area protection mechanism, high-frequency components of key areas of vehicle damage (such as collision points) are allocated independent code streams and more significant coefficients are reserved, while high-frequency components of non-key areas are selectively truncated according to rules, and a non-uniform quantization table is constructed, with a 40% reduction in quantization step size for edge gradient areas (detected by Sobel operator), ensuring clear and identifiable vehicle damage details, and ultimately making the vehicle damage detail SSIM index ≥0.95, meeting the accuracy requirements of damage determination for vehicle insurance claims.

[0063] (5) Compatibility optimization: The final output of the present disclosure is a standard PNG format image, which can directly interface with the image analysis interface of existing claims systems without the need for additional development of adaptation modules; at the same time, based on NPU parallel computing (wavelet transform speed is improved by 3 times) and asynchronous task scheduling (UI stuttering rate is reduced by 90%), 1080P image full-process processing ≤150ms, 4K image ≤300ms are realized on Snapdragon 7 Gen3 and other mid-end models, compatible with more than 95% of mainstream mobile devices.

[0064] (6) Network adaptability enhancement: The present disclosure uses dynamic quantization step size and smooth transition mechanism to ensure that there is no sudden change in image quality when the network type is switched or the bandwidth fluctuates; through three priority transmission queues combined with redundant coding, vehicle damage details can still be completely restored under a 5% packet loss rate, and it can adapt to complex network environments such as urban, suburban, and underground garages.

[0065] It should be noted that, for the foregoing method embodiments, in order to simply describe, they are all expressed as a series of action combinations, but those skilled in the art should know that the present disclosure is not limited by the action sequence described, because according to the present disclosure, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present disclosure.

[0066] The above is the introduction of the method embodiment, and the following will further illustrate the scheme of the present disclosure through the device embodiment.

[0067] Figure 3 A block diagram of an image efficient compression and fast transmission device in vehicle damage claims according to an embodiment of the present disclosure is shown. As shown in Figure 3 The device 300 includes: An image acquisition module 310 for acquiring vehicle damage scene original images through a mobile device with image acquisition function; A vehicle damage area identification and positioning module 320 for identifying and positioning the key areas of vehicle damage in the vehicle damage scene original images using a multi-modal identification method to obtain the position coordinates and area range information of the key areas of vehicle damage. The image hierarchical compression module 330 is configured to perform hierarchical compression processing on the vehicle damage scene image based on the position coordinates and the area range information of the vehicle damage key area. The image adaptive quantization module 340 is configured to perform adaptive quantization on the hierarchical compressed image data according to the real-time monitored network environment. The image data transmission module 350 is configured to upload the quantized image data to the server after processing based on a preset transmission optimization protocol, so that the server decodes and restores the image data. The standard format image generation module 360 is configured to receive the decoded and restored standard format image returned by the server.

[0068] In the technical solution of the present disclosure, the acquisition, storage and application of user personal information comply with relevant laws and regulations and do not violate public order and good customs.

[0069] According to the embodiments of the present disclosure, the present disclosure further provides an electronic device, a readable storage medium and a computer program product.

[0070] Figure 4 A schematic block diagram of an electronic device 400 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit implementations of the present disclosure described and / or claimed in this document.

[0071] The electronic device 400 includes a computing unit 401 that can perform various appropriate actions and processes in accordance with a computer program stored in a ROM 402 or a computer program loaded from the storage unit 408 into a RAM 403. In the RAM 403, various programs and data required for the operation of the electronic device 400 can also be stored. The computing unit 401, the ROM 402, and the RAM 403 are connected to each other through a bus 404. An I / O interface 405 is also connected to the bus 404.

[0072] A plurality of components in the electronic device 400 are connected to the I / O interface 405, including: an input unit 406, such as a keyboard, a mouse, etc.; an output unit 407, such as various types of displays, speakers, etc.; a storage unit 408, such as a magnetic disk, an optical disk, etc.; and a communication unit 409, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 409 allows the electronic device 400 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0073] The computing unit 401 can be various general and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 401 performs various methods and processes described above, such as the method 100. For example, in some embodiments, the method 100 can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 408. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 400 via the ROM 402 and / or the communication unit 409. When the computer program is loaded onto the RAM 403 and executed by the computing unit 401, one or more steps of the method 100 described above can be performed. Alternatively, in other embodiments, the computing unit 401 can be configured to perform the method 100 by any other appropriate means, such as by means of firmware.

[0074] Various implementations of the systems and techniques described above herein can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0075] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces a means for implementing the functions / acts specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, or entirely on a remote machine or server.

[0076] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0077] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0078] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0079] The computer system can include clients and servers. This relationship can be. The servers are typically remote from the clients with the interactions between them occurring over a communication network. The relationship between a client and a server is one of client-server. The server can be a cloud server, a server of a distributed system, or a server incorporating a blockchain.

[0080] It should be understood that the various forms of flow shown above can be re-ordered, added to, or deleted from without departing from the scope of the present disclosure. For example, the steps recited in the present disclosure can be executed in parallel, in series, or in a different order, as long as the desired results of the present disclosure are achieved, and are not limited herein.

[0081] The specific embodiments described above are not intended to be limiting, and persons skilled in the art will appreciate that various modifications, combinations, sub-combinations and alternatives can be made to the specific embodiments without departing from the spirit and scope of the disclosure. Any further modifications, equivalents, alternatives, and / or improvements made to the specific embodiments described herein are intended to fall within the scope of the disclosure.

Claims

1. A method for efficient image compression and fast transmission in vehicle damage claims, characterized in that, include: Original images of the vehicle damage scene were obtained using a mobile device with image acquisition capabilities. Multimodal recognition is used to identify and locate key areas of vehicle damage in the original images of the vehicle damage scene, and the location coordinates and area range information of the key areas of vehicle damage are obtained. Based on the location coordinates and area information of the key areas of vehicle damage, the vehicle damage scene image is subjected to layered compression processing. Adaptive quantization is performed on the layered compressed image data based on the real-time monitored network environment. The quantized image data is processed according to a preset transmission optimization protocol and then uploaded to the server so that the server can decode and restore the image data. Receive the decoded and restored standard format image returned by the server.

2. The method according to claim 1, characterized in that, A multimodal recognition method is used to identify and locate key areas of vehicle damage in the original images of the vehicle damage scene, obtaining the location coordinates and area range information of the key areas of vehicle damage, including: The original image of the vehicle damage scene is input into a pre-trained deep target detection model. Local features are extracted through CNN and multi-scale features are fused by FPN to output the preliminary location coordinates of the key areas of vehicle damage. The vehicle body outline is located by edge detection algorithm, and the edge features of key areas of vehicle damage are enhanced by combining morphological dilation and erosion operations. The collision point is located by detecting the body lines through Hough transform. By integrating the preliminary location coordinates, edge features, and collision point information of the critical vehicle damage area, the precise location coordinates and area range parameters of the critical vehicle damage area are output.

3. The method according to claim 1, characterized in that, The layered compression processing of the vehicle damage scene image based on the location coordinates and area information of the key damaged area includes: The vehicle damage scene image is decomposed into low-frequency approximation components and multiple high-frequency detail components using a biorthogonal wavelet basis. The high-frequency detail components include high-frequency detail components corresponding to the key areas of vehicle damage and high-frequency detail components corresponding to non-key areas of vehicle damage. Based on the location coordinates and area information of the key area of ​​vehicle damage, the low-frequency approximate component and multiple high-frequency detail components are subjected to layered compression processing.

4. The method according to claim 3, characterized in that, Based on the location coordinates and area information of the critical vehicle damage area, the low-frequency approximate component and the high-frequency detail component are subjected to layered compression processing, including: A lossless compression strategy is applied to the low-frequency approximate components to preserve basic image information; Using the location coordinates and area information of the key vehicle damage area, all significant coefficients of the high-frequency detail components corresponding to the key vehicle damage area are retained and an independent code stream is assigned; The significance coefficients of high-frequency detail components corresponding to non-vehicle damage critical areas are selectively truncated.

5. The method according to claim 1, characterized in that, The adaptive quantization of the layered compressed image data based on the real-time monitored network environment includes: Network environment parameters are obtained in real time through the mobile terminal network interface, wherein the network environment parameters include at least network type, real-time bandwidth and latency jitter; Based on the real-time acquired network environment parameters and the preset mapping relationship between network quality and quantization step size, the corresponding quantization step size is automatically switched. When the network type changes or bandwidth fluctuations exceed a preset fluctuation threshold or latency jitter exceeds a preset time threshold, a gradual adjustment strategy is adopted to dynamically adjust the quantization step size.

6. The method according to claim 3, characterized in that, The quantized image data is processed according to a preset transmission optimization protocol and then uploaded to the server, including: The quantized image data is intelligently segmented based on the network's maximum transmission unit. The fragmented image data is divided into three priority data packets: high-frequency detail components corresponding to the critical vehicle damage area are high-priority data packets, low-frequency approximation components are medium-priority data packets, and high-frequency detail components corresponding to non-critical vehicle damage areas are low-priority data packets. According to the pre-set priority scheduling strategy, data is scheduled and transmitted to the server through a three-level priority queue set up in the transport layer.

7. The method according to claim 6, characterized in that, According to a pre-defined priority scheduling policy, data is scheduled and transmitted to the server through a three-level priority queue at the transport layer, including: According to the pre-set priority scheduling strategy, high-priority data packets are inserted at the head of the transmission queue, medium-priority data packets are inserted in the middle of the queue in the order of arrival, and low-priority data packets are inserted at the tail of the transmission queue. The BBR algorithm is used to dynamically adjust the transmission rate. When network congestion occurs, high-priority data packets still maintain a minimum 50% bandwidth reservation.

8. The method according to claim 7, characterized in that, The method further includes: The high-priority data packets are subjected to forward error correction coding to generate redundant data packets with a preset ratio, wherein the redundant data packets inherit the priority marker of their corresponding original data packets; When the network packet loss rate exceeds a preset threshold, a mechanism for prioritizing the transmission of redundant data packets is triggered.

9. The method according to claim 1, characterized in that, The server decodes and restores the image data, including: Context-based adaptive binary arithmetic coding is used to decode image data; Wavelet reconstruction is performed on the decoded image data, and inverse transform is performed through the prediction step and the update step; The inversely transformed image data is encapsulated into a standard format according to the PNG standard.

10. A device for efficient image compression and rapid transmission in vehicle damage claims, characterized in that, include: The image acquisition module is used to acquire original images of the vehicle damage scene through a mobile device with image acquisition capabilities; The vehicle damage area identification and positioning module is used to identify and locate the key vehicle damage areas in the original image of the vehicle damage scene using a multimodal recognition method, and obtain the location coordinates and area range information of the key vehicle damage areas; The image layer compression module is used to perform layer compression processing on the vehicle damage scene image based on the location coordinates and area range information of the key vehicle damage area; The image adaptive quantization module is used to adaptively quantize the layered compressed image data according to the real-time monitored network environment. The image data transmission module is used to process the quantized image data according to a preset transmission optimization protocol and then upload it to the server so that the server can decode and restore the image data. The standard format image generation module is used to receive the decoded and restored standard format image returned by the server.

Images