Video monitoring data transmission and storage method based on narrow bandwidth

Abstract

The invention relates to the technical field of video compression, in particular to a narrow-bandwidth-based video monitoring data transmission and storage method, which comprises the following steps of: for an input video monitoring data frame, calculating a difference absolute value sum between a current frame and a reference frame; according to the method, the pixel difference of the current frame and the reference frame is calculated, and the average amplitude of the motion vector field is fused to form a comprehensive index of the dynamic degree of the quantized content, so that the image group structure is not fixed or periodic any more, and the dynamic degree of the quantized content can be calculated according to the score of scene activity and the change intensity. When a monitoring picture is static, a super-long image group is established to limit a compression code rate, when the picture is suddenly changed, the super-long image group is quickly switched to a short image group to ensure instant refreshing and definition of key information, meanwhile, non-uniform redistribution is performed on limited total code rate budget, bit resources can be intelligently inclined to frames with violent content change and large information amount, and the real-time refreshing and definition of the key information are ensured. Therefore, the subjective visual quality of the key dynamic moments is improved under the narrow bandwidth.

Description

Technical Field

[0001] This invention relates to the field of video compression technology, and in particular to a method for transmitting and storing video surveillance data based on narrow bandwidth. Background Technology

[0002] The field of video compression technology uses specific processing algorithms to significantly reduce the amount of data required to represent video signals while maintaining visual quality perceptible to the human eye as much as possible.

[0003] Current video compression technologies typically rely on a set of preset, relatively fixed encoding parameters to handle all types of video content. For example, using a fixed-period frame group structure can waste valuable bandwidth and storage space when the monitored image is static for a long time due to the periodic insertion of unnecessary keyframes. Conversely, when a sudden event causes a dramatic change in the image, the fixed long frame group structure cannot insert new keyframes in time, resulting in a decrease in the encoding quality of subsequent predicted frames. Key dynamic details are blurred or severely covered by artifacts until the next predetermined keyframe arrives. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a video surveillance data transmission and storage method based on narrow bandwidth.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a video surveillance data transmission and storage method based on narrow bandwidth, comprising the following steps: For the input video surveillance data frames, calculate the sum of absolute differences between the current frame and the reference frame, simultaneously extract the average amplitude of the motion vector field, and perform calculations between the sum of absolute differences and the average amplitude of the motion vector field to obtain the inter-frame dynamic quantization value. Summarize the inter-frame dynamic quantization values ​​to obtain the scene activity score. The scene activity score is compared with a preset low activity threshold T1 and an activity mutation threshold T2 to generate an image group structure instruction. An adaptive image group structure is then established based on the image group structure instruction. Based on the adaptive image group structure, the scene activity score is used to redistribute the total bitrate budget under narrow bandwidth conditions within the image group, a frame-level bitrate allocation scheme is established, and the coding unit is predicted, transformed and quantized according to the frame-level bitrate allocation scheme to obtain the prediction residual spectrum data. The energy of the predicted residual spectrum data is normalized, the KL divergence between the probability distribution and the uniform distribution is calculated, and the spatial frequency spectrum divergence value is obtained. The spatial frequency spectrum divergence value is compared with the artifact threshold and the noise threshold to generate an adaptive loop filter configuration.

[0006] Preferably, the steps for obtaining the scene activity score are as follows: For the grayscale values ​​of the corresponding pixels of the input video surveillance data frame and the reference frame, the grayscale difference is calculated pixel by pixel and the absolute value is taken. Then, the average value is calculated according to the number of pixels to obtain the average pixel difference. At the same time, the horizontal and vertical components of the motion vector are extracted, the amplitude is calculated, and the average value is taken to form the average amplitude of the motion vector field. The average pixel difference and the average amplitude of the motion vector field are generated. Based on the average pixel difference and the average amplitude of the motion vector field, normalization processing is performed, and inter-frame dynamic quantization values ​​are calculated simultaneously. Based on the inter-frame dynamic quantization value, the inter-frame dynamic quantization values ​​of consecutive frames within the time window are accumulated and divided by the number of frames in the window to obtain the time average index. The rate of change of the dynamic quantization values ​​between adjacent frames is detected. When the rate of change exceeds the activity detection threshold, the window is re-initialized and a scene activity score is generated.

[0007] Preferably, the step of obtaining the image group structure instruction is as follows: Based on the scene activity score, the current scene activity score and the scene activity score at the previous moment are read, and the absolute value of the difference between the two is calculated. The current scene activity score is compared with the low activity threshold T1, and the absolute value of the difference is compared with the activity mutation threshold T2. The control labels are mapped according to three rules: less than T1 and the absolute value of the difference is less than T2, the absolute value of the difference is greater than or equal to T2, and the rest are generated to generate image group structure instructions.

[0008] Preferably, the step of obtaining the adaptive image group structure is as follows: According to the image group structure instruction, when the control label is static and stable, an ultra-long GOP structure is selected; when the control label is abrupt, a short GOP structure is selected; and when the control label is stable and dynamic, a regular GOP structure is selected. The group frame length, key frame interval, and reference frame level of the selected GOP structure in the video surveillance data stream are recorded to generate the GOP structure selection result of the video surveillance data stream. Based on the GOP structure selection result of the video surveillance data stream, the group frame length, key frame interval and reference frame level are written during the encoding initialization stage of the video surveillance data stream. They are applied to subsequent frame groups in chronological order and a switching boundary is established. Parameters are reset and the buffer queue is refreshed at the cross-segment switching points to generate an adaptive image group structure.

[0009] Preferably, the step of obtaining the frame-level bitrate allocation scheme is as follows: Based on the adaptive image group structure, the frame order position, frame type and reference level within the image group are read. Combined with the scene activity score and total bitrate budget, frame quotas are allocated according to frame order position and reference level, and the frame quotas within the keyframe interval are recorded to generate a frame-level bitrate allocation scheme.

[0010] Preferably, the step of obtaining the predicted residual spectrum data is as follows: According to the frame-level bit rate allocation scheme, the target bits of the coding unit are divided according to the frame quota. A reference pixel buffer is established for each coding unit and a prediction signal is generated. The difference block between the original pixel and the prediction signal is calculated. After block-domain frequency domain transformation, the signal is quantized according to the quantization step size to generate the prediction residual block coefficient matrix. Based on the predicted residual block coefficient matrix, the coefficients of the same frequency index are merged in block order, the amplitude distribution and energy proportion of each frequency index are statistically analyzed, a frequency band division table is constructed, and the amplitude statistics and energy statistics of each frequency band are output to generate the predicted residual spectrum data.

[0011] Preferably, the step of obtaining the spatial frequency spectrum divergence value is as follows: Based on the predicted residual spectrum data, the coefficients are divided into multiple frequency bands according to the frequency index. The sum of the square magnitudes of all coefficients in each frequency band is divided by the total number of coefficients in the frequency band. The average energy of each frequency band is calculated, the total energy of the entire frequency band is calculated, and the average energy of each frequency band is divided by the total energy of the entire frequency band to obtain the energy proportion sequence, thereby generating a normalized frequency band energy distribution. The spatial frequency spectrum divergence value is calculated based on the normalized frequency band energy distribution.

[0012] Preferably, the step of obtaining the adaptive loop filter configuration is as follows: Based on the spatial frequency spectrum divergence value, compare the spatial frequency spectrum divergence value with the artifact threshold and the noise threshold. If the spatial frequency spectrum divergence value exceeds the artifact threshold, select strong filtering and enable edge enhancement mode. If the spatial frequency spectrum divergence value is between the two thresholds, select medium filtering and enable smoothing mode. If the spatial frequency spectrum divergence value is lower than the noise threshold, turn off the filtering module and generate an adaptive loop filtering configuration.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention calculates the pixel differences between the current frame and the reference frame and fuses the average amplitude of the motion vector field to form a comprehensive index that quantifies the dynamic degree of the content. This makes the image group structure no longer fixed or periodic, but can be based on the level and intensity of scene activity scores. When the monitored scene is static, ultra-long image groups are built to maximize the compression rate. When the scene changes abruptly, it quickly switches to short image groups to ensure the instant refresh and clarity of key information. At the same time, the limited total bitrate budget is non-uniformly redistributed to ensure that bit resources are intelligently tilted to frames with drastic content changes and large amounts of information. This improves the subjective visual quality of key dynamic moments under narrow bandwidth. Furthermore, at the end of the encoding process, the deviation of the predicted residual spectrum energy distribution from the ideal uniform distribution is quantified to identify and distinguish block artifacts and random noise. This drives the adaptive selection of loop filter configuration to achieve targeted repair of compression distortion and avoid over-smoothing of high-quality areas. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the steps of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0016] Please see Figure 1 This invention provides a technical solution: a method for transmitting and storing video surveillance data based on narrow bandwidth, comprising the following steps: For the input video surveillance data frames, calculate the sum of absolute differences between the current frame and the reference frame, simultaneously extract the average amplitude of the motion vector field, and perform calculations between the sum of absolute differences and the average amplitude of the motion vector field to obtain the inter-frame dynamic quantization value. Summarize the inter-frame dynamic quantization values ​​to obtain the scene activity score. The scene activity score is compared with the preset low activity threshold T1 and activity mutation threshold T2 to generate image group structure instructions. Based on the image group structure instructions, an adaptive image group structure is established. Based on the adaptive image group structure, the scene activity score is used to redistribute the total bitrate budget under narrow bandwidth conditions within the image group, establish a frame-level bitrate allocation scheme, and perform prediction, transformation and quantization on the coding unit according to the frame-level bitrate allocation scheme to obtain prediction residual spectrum data. The energy of the predicted residual spectrum data is normalized, the KL divergence between the probability distribution and the uniform distribution is calculated, and the spatial frequency spectrum divergence value is obtained. The spatial frequency spectrum divergence value is compared with the artifact threshold and the noise threshold to generate an adaptive loop filter configuration.

[0017] The steps to obtain the scene activity score are as follows: For the grayscale values ​​of the corresponding pixels of the input video surveillance data frame and the reference frame, the grayscale difference is calculated pixel by pixel and the absolute value is taken. Then, the average value is calculated according to the number of pixels to obtain the average pixel difference. At the same time, the horizontal and vertical components of the motion vector are extracted, the amplitude is calculated, and the average value is taken to form the average amplitude of the motion vector field. The average pixel difference and the average amplitude of the motion vector field are generated. Normalization is performed based on the average pixel difference and the average amplitude of the motion vector field. Simultaneously, the inter-frame dynamic quantization value is calculated using the following formula: ; in, For inter-frame dynamic quantization values, This is the normalized average pixel difference. This represents the normalized average amplitude of the motion vector field. Contribution coefficient to pixel difference normalization Contribution coefficients for normalization of the motion vector field. This is the synergistic effect adjustment coefficient. and Using formulas respectively , Calculation, where This represents the theoretical maximum value of the pixel grayscale difference. A is the upper limit of the motion vector amplitude, A is the average value of the absolute value of the pixel grayscale difference between the current frame and the reference frame, and B is the average value of the amplitude of each motion vector in the motion vector field of the current frame. Based on the inter-frame dynamic quantization value, the inter-frame dynamic quantization values ​​of consecutive frames within the time window are accumulated and divided by the number of frames in the window to obtain the time average index. The rate of change of the dynamic quantization value between adjacent frames is detected. When the rate of change exceeds the activity detection threshold, the window is re-initialized and a scene activity score is generated.

[0018] Specifically, for the corresponding pixel grayscale values ​​of the input video surveillance data frame and the reference frame, the current processing frame is first locked. With reference frame as the benchmark The image size of both frames is Pixels, then iterate through each pixel in the image. ,in The range is from 1 to , The range is from 1 to Extract two frames at the same coordinates 8-bit grayscale value and Calculate the absolute value of the difference between the two. Sum the absolute values ​​of the differences between all pixels, then divide by the total number of pixels. The average pixel difference A is obtained, and at the same time, the current frame is... The motion estimation algorithm is executed, specifically by dividing the current frame into several non-overlapping 16×16 pixel macroblocks. For each macroblock, in the reference frame... Within a preset search window, such as a square area of ​​±32 pixels centered on the macroblock, a diamond search algorithm is used to find the best matching block. The matching criterion is to minimize the sum of absolute differences (SAD), which is calculated by summing the absolute values ​​of the pixel grayscale differences between the current macroblock and the candidate blocks in the search window. The positional offset between the candidate block with the smallest SAD value and the current macroblock is the motion vector of that macroblock. After calculating the motion vectors of all macroblocks in the current frame, the magnitude of each motion vector is calculated. The motion vector amplitudes of all macroblocks are summed and then divided by the total number of macroblocks to obtain the average amplitude of the motion vector field B. Finally, the pixel difference average value and the average amplitude of the motion vector field are generated.

[0019] In the formula for calculating inter-frame dynamic quantization values, the first part of the formula... The first part is linear, directly reflecting the contribution of two independent dynamic factors—pixel brightness changes and object motion—to scene activity. The second part... This is the synergistic effect term, used to capture the enhancement effect when pixel changes and motion occur simultaneously. For example, when a richly textured object moves quickly, the resulting scene dynamics are much greater than when a solid-color object moves at the same speed, or when scene lighting changes but no object moves. The square root of this term... Used to smooth the product effect of two factors, while the natural logarithm It provides approximately linear growth when the product of the two is small, and suppresses it when the product is large to prevent numerical explosion, thus making the response of the dynamic quantization value more robust and in line with the characteristics of human visual perception.

[0020] This is the normalized average pixel difference. This value is obtained by first calculating the average of the absolute values ​​of the pixel grayscale differences between the current frame and the reference frame, A, in the previous step. Considering that video data typically uses 8-bit depth to represent grayscale, and the grayscale value of each pixel ranges from 0 to 255, the theoretical maximum value of the grayscale difference between two pixels is... The value is 255, and the normalization formula is: This formula compares the actual measured average pixel difference with the theoretical maximum difference to obtain a ratio between 0 and 1. For example, in an indoor surveillance video, if a frame experiences slight flickering light, and the average grayscale difference A calculated pixel-by-pixel is 5.1, then the normalized average pixel difference... The calculation process is as follows .

[0021] This is the normalized average amplitude of the motion vector field, derived from the average amplitude B of each motion vector in the current frame's motion vector field calculated in the previous step. It represents the upper limit of the motion vector amplitude used for normalization. Determined by the motion search range parameter of the video encoder, in a typical 1080p video encoding configuration, the motion search range is often set to ±32 pixels. In this case, theoretically, the maximum horizontal and vertical components of a motion vector are both 32, and its maximum amplitude is [value missing]. Therefore, set The normalization calculation formula is: For example, in a parking lot surveillance video, a car slowly drives by, and the calculated average motion vector amplitude B is 4.8 pixels. Then, the normalized average amplitude of the motion vector field is... The calculation process is as follows .

[0022] The contribution coefficient for pixel difference normalization is used to adjust the weight of pixel content changes in the final inter-frame dynamic quantization value. Its value is set based on the analysis results of a large number of different types of surveillance video scenes (such as indoor static, outdoor traffic, and low-light nighttime). The specific setting process involves preparing a test video set containing 500 hours of various surveillance recordings, manually annotating the video set, and marking three types of scene segments: "lighting change-dominated," "object movement-dominated," and "mixed dynamic." For segments dominated by "lighting change," adjustments are made... The value makes the calculated The value had the highest Pearson correlation coefficient with the observed visual change amplitude. While keeping other coefficients constant, the results were obtained by statistically averaging the optimized results of all such segments. Based on the baseline values, experiments show that the model performs optimally when pixel changes contribute slightly less to scene dynamics than motion information. Therefore, the final setting is... The value is 0.4.

[0023] The contribution coefficient for normalizing the motion vector field determines the proportion of motion information in the inter-frame dynamic quantization values, and its setting process is similar to... Similarly, using the same 500-hour annotated video set, but this time focusing on scene segments labeled as "object-dominated," adjustments were made to these segments. The value of is to maximize. The correlation between the value and the dynamic activity reflected by the speed, quantity, and size of moving objects in the video is crucial. In video surveillance applications, object movement is generally considered a more important event than changes in lighting, therefore its weight should be relatively high. Through systematic testing and statistical analysis of all "object movement-dominated" segments, it was found that when... When the value is set to 0.5, the model achieves the best balance between sensitivity to motion events and quantification accuracy; therefore, it is ultimately determined to be [value missing]. The value is 0.5.

[0024] This is the synergistic effect adjustment coefficient, used to control the strength of the interaction between pixel changes and motion vectors, i.e., adjusting the contribution of the nonlinear term to the total dynamic quantization value. Its value is set primarily for "hybrid dynamic" video clips, where lighting changes and object motion coexist. The goal is to make The value accurately reflects this superposition effect, rather than simply being equal to the sum of two independent effects. The setting process is as follows: and Based on the established conditions, traverse Given possible values ​​(e.g., from 0.1 to 1.0, with a step size of 0.05), calculate the model's output on the "mixed dynamics" segment for each value. The value sequence is compared with the manually rated scene complexity score sequence, and the sequence that best represents the consistency between the two is selected. The value, according to analysis, is a moderate The value can effectively amplify the signal of real dynamic events while suppressing noise, and ultimately determine... The value is 0.3.

[0025] Calculations based on parameters: In a specific scenario, the average value A of the absolute value of the pixel grayscale difference between the current frame and the reference frame, calculated through the aforementioned steps, is 8.2, and the average value B of the amplitude of each motion vector in the motion vector field of the current frame is 6.5. Based on the settings in the parameter acquisition steps, the theoretical maximum value of the pixel grayscale difference is... The upper limit of the motion vector amplitude is 255. The contribution coefficient of pixel difference normalization is 45.25. The normalized contribution coefficient of the motion vector field is 0.4. The synergistic effect moderating coefficient is 0.5. The value is 0.3. First, calculate the normalized value: ; ; Substitute the above values ​​into the inter-frame dynamic quantization values. In the calculation formula: ; ; The result indicates that the inter-frame dynamic quantization value of the current frame is 0.0849. The larger the value, the more intense the scene activity. The result of 0.0849 indicates that the scene is in a relatively stable but slightly dynamic state, such as a person moving around in a small area at their seat in an office scene, or the leaves swaying slightly in an outdoor scene.

[0026] Based on the inter-frame dynamic quantization value, a first-in-first-out queue of length N frames is maintained as a time window. For example, for a video stream with a frame rate of 25fps, N is set to 50, which is a 2-second time window. After each new frame is processed, its inter-frame dynamic quantization value is calculated. Then, push the value to the tail of the queue, and remove the oldest value from the head of the queue. Then calculate all 50 in the queue. The average of the values, which is the time average index at the current moment, effectively smooths out the noise caused by single-frame noise or instantaneous disturbances. The value fluctuates drastically. At the same time, a parallel rate of change detection process is executed to identify sudden scene changes. This process calculates the dynamic quantization value that has just entered the queue. The value of the immediately preceding frame absolute difference between This difference is then compared to a dynamic activity detection threshold, which is set by continuously calculating the standard deviation of the dynamic quantization differences between all adjacent frames over the past M frames (e.g., M=250, i.e., 10 seconds). And set the activity detection threshold to k times that standard deviation, for example, k=3, which is the threshold. When the calculated difference When the dynamic threshold is exceeded, for example, when the monitoring screen suddenly switches from an empty corridor to a crowded conference room, If the value jumps from 0.05 to 0.6, and the difference far exceeds the threshold calculated based on the preceding stable scenario, a scenario abrupt change is determined. In this case, the time window queue is immediately cleared, and the current time window is used. The values ​​are refilled in the queue for subsequent frames. The value will be accumulated and averaged based on this new starting point. This avoids the statistical characteristics of the old scene from contaminating the activity assessment of the new scene. Finally, the time average index output at each time point is adopted as the scene activity score at that time.

[0027] The steps for obtaining image group structure instructions are as follows: Based on the scene activity score, read the current scene activity score and the scene activity score at the previous moment, calculate the absolute value of the difference between the two, compare the current scene activity score with the low activity threshold T1, and compare the absolute value of the difference with the activity mutation threshold T2. Map control labels according to three rules: less than T1 and the absolute value of the difference is less than T2, the absolute value of the difference is greater than or equal to T2, and the rest, and generate image group structure instructions.

[0028] Specifically, based on the scene activity score, the scene activity score generated at the current time point is read from the internal memory. And the scene activity score at the next adjacent time point. Then calculate the absolute value of the difference between the two fractions. Next The activity threshold T1 is compared with a preset low activity threshold T1, and the absolute difference D is compared with the activity mutation threshold T2. The setting process for these two thresholds is as follows: First, the low activity threshold T1 is determined by offline analysis of a video database containing over 1000 hours of various monitoring scenarios (such as nighttime warehouses, offices, busy streets, parking lots, etc.). The scene activity score is calculated frame-by-frame for all videos in the database, forming a large score sample set. Then, statistical distribution analysis is performed on this sample set, and its probability density function graph is plotted. Typically, the scores corresponding to low activity scenarios (such as empty corridors) will be concentrated in the low-value region of the distribution. The score corresponding to a cumulative distribution function value of 0.1 is selected as T1. For example, after statistical calculation, it is found that 10% of the scene activity scores are below 0.04, so T1 = 0.04 is set. Second, the activity mutation threshold T2 is determined based on the analysis of the first-order difference sequence of scene activity scores, that is, the analysis of the first-order difference sequence of all videos in the aforementioned database. Statistical analysis of the values ​​reveals that most scenes with stable changes have small differences, while abrupt events such as scene transitions, violent camera shake, or explosions produce large differences, forming long tails or outliers in the distribution. The 99th percentile of this difference distribution is selected as T2. For example, if the absolute value of 99% of the inter-frame score changes is less than 0.35, then T2 is set to 0.35. After setting the threshold, judgment and mapping are performed according to the following three types of rules: First, if the current scene activity score... If the difference is less than 0.04 and the absolute value of the difference D is less than 0.35, the current scene is determined to be static and stable, and a control label with a value of "0" is generated. In the second category, if the absolute value of the difference D is greater than or equal to 0.35, regardless of the current scene activity score, it is determined to be a sudden change in the scene, and a control label with a value of "1" is generated. The third category excludes all other cases, such as... If D is 0.15 and D is 0.02, it is determined to be a stable dynamic scene, and a control label with a value of "2" is generated. Finally, the generated control label "0", "1" or "2" is used as the image group structure instruction.

[0029] The steps for obtaining the adaptive image group structure are as follows: Based on the image group structure instruction, the ultra-long GOP structure is selected when the control label is static and stable, the short GOP structure is selected when the control label is abrupt, and the regular GOP structure is selected when the control label is stationary and dynamic. The group frame length, key frame interval and reference frame level of the selected GOP structure in the video surveillance data stream are recorded to generate the GOP structure selection result of the video surveillance data stream. Based on the GOP structure selection result of the video surveillance data stream, the group frame length, key frame interval and reference frame level are written during the encoding initialization stage of the video surveillance data stream. They are applied to subsequent frame groups in chronological order and a switching boundary is established. Parameters are reset and the buffer queue is refreshed at the cross-segment switching points to generate an adaptive image group structure.

[0030] Specifically, based on the group of pictures (GOP) structure instructions, i.e., upon receiving control labels "0", "1", or "2" representing the scene state, a predefined mapping query operation is executed. This operation searches for the corresponding group of pictures (GOP) structure parameters in an internal configuration table. This configuration table is pre-set during system initialization according to the specific requirements of narrow bandwidth transmission and the characteristics of video coding standards (such as H.264 or H.265). When the received control label is "0", corresponding to a static and stable scene, an ultra-long GOP structure is selected. Its specific parameter configuration is as follows: the group frame length is set to 300 frames, the key frame (I-frame) interval is also set to 300 frames, and the reference frame level adopts a layered B-frame structure with a depth of 5 layers. For example, a typical prediction structure is IBBBPBBBP..., where the number of B-frames between P-frames is relatively large and the shape is... A multi-level reference relationship is established. When the received control tag is "1", corresponding to a sudden change scene, a short GOP structure is selected. At this time, the group frame length is set to 15 frames, the key frame interval is also 15 frames, and the reference frame level is simplified to 1 level. A simple prediction structure such as IPPP is used, and B-frames are used little or no. When the received control tag is "2", corresponding to a stable dynamic scene, a regular GOP structure is selected. The group frame length is set to 60 frames, the key frame interval is 60 frames, and the reference frame level is set to a 3-level layered B-frame structure. This is a configuration that balances compression efficiency and random access latency. After determining the specific GOP structure parameters, the three core parameters of the selected group frame length, key frame interval, and reference frame level are packaged into a data structure. This data structure is the GOP structure selection result of the video surveillance data stream.

[0031] Based on the GOP structure selection result of the video surveillance data stream, when the video encoding process is about to start processing a new group of images (GOP), the group frame length, keyframe interval, and reference frame level parameters contained in the result are written into the core parameter register or configuration data structure of the video encoder. This write operation marks the beginning of a new adaptive GOP segment. The encoder then organizes the subsequent frame encoding sequence according to these new parameters. For example, if the received parameters are a group frame length of 300 and a keyframe interval of 300, the encoder will encode the current frame as an I-frame and set a GOP counter that lasts for 300 frames. In the next 299 frames, based on the set 5-layer reference frame level, it determines whether each frame is a P-frame or a B-frame, and their reference relationships. This process is applied frame by frame in chronological order until the current GOP ends or is interrupted by a new instruction. Simultaneously, when transitioning from a GOP structure... At the boundary of switching to another structure, such as switching from a very long GOP of 300 frames to a short GOP of 15 frames, boundary processing logic is performed. Before encoding the first I-frame of the new GOP, the encoder state parameters are reset. This includes clearing the decoded image buffer (DPB), discarding all reference frames from the previous GOP, resetting the historical motion information used by the motion vector predictor, and refreshing any context model related to time prediction. This refresh action ensures that the new I-frame is a completely independent decode refresh point (IDR frame), and subsequent frames will not incorrectly reference images in GOPs with completely different scene content from the previous one, thus establishing a clear switching boundary. By dynamically applying parameters and managing switching boundaries in this way, a series of GOP sequences with different lengths and structures are generated in the data stream, forming a complete adaptive image group structure.

[0032] The steps for obtaining the frame-level bitrate allocation scheme are as follows: Based on the adaptive image group structure, the frame order position, frame type and reference level within the image group are read. Combined with the scene activity score and total bitrate budget, frame quotas are allocated according to frame order position and reference level, and the frame quotas within the keyframe interval are recorded to generate a frame-level bitrate allocation scheme.

[0033] Specifically, based on the adaptive group of pictures (GOP) structure, the total bitrate budget for the current GOP is first determined. This budget is calculated by dividing the system-set total target bitrate (e.g., 2 Mbps) by the video frame rate (e.g., 25 fps) to obtain bits per second, and then multiplying this by the length of the current GOP (e.g., a typical GOP is 60 frames), resulting in a total GOP bitrate of 9600. Next, an initial bit quota is set for each frame within the GOP, based on the frame type and its position in the reference hierarchy. Specifically, a basic weight table is established; for example, I-frames have a weight of 12, P-frames have a weight of 4, B-frames at reference hierarchy 0 (directly referencing I / P frames) have a weight of 2, and B-frames at reference hierarchy 1 (referencing other B-frames) have a weight of 1. Then, these basic weights are multiplied by a dynamic adjustment factor, which is determined by the scene activity score of the corresponding frame. For example, the adjustment factor can be set to... Here, S represents the scene activity score of the frame. A frame with a higher scene activity score will have a higher final weight. Then, the final weights of all frames within the GOP are summed, and the total GOP bitrate budget is divided by this total weight to obtain the number of bits per unit weight. Finally, the final weight of each frame is multiplied by this number of bits per unit weight to obtain the precise frame quota for that frame. For example, if a P-frame has a base weight of 4 and a scene activity score of 0.2, its adjustment factor is 1.1, and its final weight is 4.4. If the number of bits per unit weight is 5000 bits, then the quota for that P-frame is 22000 bits. The calculated quota for each frame within the GOP and its corresponding frame number are recorded together to form a list containing [frame number, frame quota] key-value pairs. This list is the frame-level bitrate allocation scheme.

[0034] The steps for obtaining the predicted residual spectrum data are as follows: According to the frame-level bit rate allocation scheme, the target bits of the coding unit are divided according to the frame quota. A reference pixel buffer is established for each coding unit and a prediction signal is generated. The difference block between the original pixel and the prediction signal is calculated. After block-domain frequency domain transformation, the signal is quantized according to the quantization step size to generate the prediction residual block coefficient matrix. Based on the predicted residual block coefficient matrix, the coefficients of the same frequency index are merged in block order, the amplitude distribution and energy proportion of each frequency index are statistically analyzed, a frequency band division table is constructed, and the amplitude statistics and energy statistics of each frequency band are output to generate the predicted residual spectrum data.

[0035] Specifically, according to the frame-level bitrate allocation scheme, for the frame to be encoded, its allocated frame quota is first obtained. Then, the frame is divided into multiple coding units (CUs), such as a quadtree structure from 64x64 to 8x8. The frame quota is initially allocated according to the complexity of each CU. The complexity of a CU is measured by calculating the variance of its internal pixels. The larger the variance, the more complex the content of the CU is considered, and the more bits are initially allocated. For each CU, the prediction mode decision process is initiated. If it is intra-frame coding, all predefined intra-frame prediction modes are traversed, such as the 35 modes in the H.265 standard. A prediction block is generated for each mode, and the sum of absolute differences (SAD) between the prediction block and the original pixel block is calculated. The mode with the smallest SAD is selected as the best prediction mode, and its corresponding prediction block is the prediction signal. If it is inter-frame coding, the prediction block is determined from the reference frame list. Motion estimation is performed by finding the best matching block using algorithms such as diamond search to obtain motion vectors and reference frame indices. Motion compensation is then used to generate a prediction signal. After obtaining the prediction signal, the difference between the original CU block and the prediction signal is calculated pixel by pixel to form a prediction residual block. Next, an integer discrete cosine transform (DCT) is applied to the residual block to transform it from the spatial domain to the frequency domain, resulting in a transform coefficient block containing DC and AC coefficients. Finally, based on the target number of bits allocated to the CU, the optimal quantization step size (QP) is determined through a rate-distortion optimization model. The larger the QP value, the coarser the quantization and the fewer bits are generated, and vice versa. The selected QP is used to perform scalar quantization on each coefficient in the transform coefficient block, that is, the coefficient is divided by the quantization step size value corresponding to the QP and rounded to the nearest integer. After this series of operations traverses all CUs in the frame, the set of all quantized coefficient blocks constitutes the prediction residual block coefficient matrix.

[0036] Based on the predicted residual block coefficient matrix, initiate a statistical analysis process for the full-frame coefficients. This process first initializes a set of data structures to store the statistical information for each frequency index. For an 8x8 transform block, there are 64 frequency indices, ranging from the DC component represented by (0,0) to the highest-frequency AC component represented by (7,7). Then, traverse each quantized predicted residual block in the frame in raster scan order. For each non-zero coefficient within the block, read its frequency index (u,v) within the block and its quantized magnitude, and then append this magnitude to the global statistical list corresponding to the frequency index (u,v). At the same time, accumulate the square of this magnitude into the energy accumulator corresponding to the frequency index. After traversing all non-zero coefficients in all blocks within the frame, the coefficient magnitude distribution list and total energy for each frequency index are obtained. Subsequently, group these frequency indices according to a preset frequency band division table, which is designed based on the sensitivity characteristics of the human visual system to different spatial frequencies. For example, the 64 frequency indices can be divided into 4 frequency bands. Band 1 contains the DC coefficient (0,0), Band 2 contains low-frequency AC coefficients (such as u + v <= 2), Band 3 contains medium-frequency AC coefficients (such as 2 < u + v <= 6), and Band 4 contains high-frequency AC coefficients (u + v > 6). Then, for each frequency band, sum the accumulated energy values of all frequency indices it contains to obtain the total energy of this frequency band, and汇总 all frequency band energies to obtain the total energy of the frame. Calculate the percentage of each frequency band energy in the total energy. At the same time, merge the magnitude lists of all frequency indices within the frequency band to generate the magnitude distribution histogram of each frequency band. Finally, organize the energy value, energy percentage, and magnitude statistical data of each frequency band into a structured data set, and this data set is the predicted residual spectrum data.

[0037] The steps to obtain the spatial frequency spectrum divergence value are as follows: According to the predicted residual spectrum data, divide the coefficients into multiple frequency bands according to the frequency index. Sum the squared magnitudes of all coefficients in each frequency band and divide by the total number of coefficients in the frequency band to calculate the average energy of each frequency band. Statistically sum the total energy of the full frequency band. Divide the average energy of each frequency band by the total energy of the full frequency band to obtain the energy percentage sequence and generate the normalized frequency band energy distribution; According to the normalized frequency band energy distribution, calculate the spatial frequency spectrum divergence value. The calculation formula is: ; Where, is the spatial frequency spectrum divergence value, is the energy percentage of the k-th frequency band, calculated as the ratio of the energy of the k-th frequency band to the total energy of the full frequency band, is the uniform percentage of the k-th frequency band, with a value of , is the number of frequency bands, The visual perception weights for the k-th frequency band are set based on the human eye's sensitivity to different frequencies. This is the balance coefficient.

[0038] Specifically, based on the predicted residual spectrum data, which contains the quantization coefficient matrix of all predicted residual blocks within a frame, all possible frequency indices are first grouped according to a pre-defined frequency band division table. This table is based on modeling the contrast sensitivity function (CSF) of the human visual system (HVS). Specifically, through numerous psychophysical experiments, the recognition threshold of the human eye for sinusoidal grating patterns at different spatial frequencies (corresponding to different frequency indices after DCT transformation) is measured, resulting in a curve describing sensitivity. This curve typically peaks in the low-to-mid-frequency region and decreases in the DC and extremely high-frequency regions. Based on the shape of this curve, an 8x8... The 64 frequency indices of the DCT block are divided into 8 frequency bands of unequal visual importance. For example, band 1 contains only the DC component (0,0), band 2 contains the lowest frequency AC components, bands 3 to 6 gradually cover the mid-frequency region, and bands 7 and 8 contain the highest frequency AC components. After the division, all prediction residual coefficients of the current frame are traversed, and the squared magnitude of each coefficient is accumulated into the total energy counter of its respective frequency band. After processing all coefficients, the total energy of each frequency band is obtained. Next, the total energy of all frequency bands is added together to obtain the total energy of the current frame across all frequency bands. Finally, the total energy of each frequency band is... Divide by the total energy of the entire frequency band The energy percentage of this frequency band was calculated. The energy proportion of all frequency bands Arrange them in order into a vector to generate a normalized frequency band energy distribution.

[0039] The formula for calculating the spatial frequency spectrum divergence combines two different dimensions of distortion measurement. The first part of the formula is a weighted KL divergence term, which measures the actual distribution of residual energy. With completely unstructured uniform noise distribution The difference lies in the energy concentration of a highly structured natural image residual within a few frequency bands, resulting in a large KL divergence value. Severe block artifacts also concentrate energy at specific frequencies corresponding to block boundaries, similarly increasing the KL divergence value. The second part is the maximum value of the high-frequency band weighted energy. This addresses high-frequency distortions such as mosquito noise or ringing, which are highly sensitive to human vision. These distortions introduce abnormal energy spikes in the high-frequency region, which are then addressed through a balance coefficient. It can dynamically adjust the attention given to uneven global energy distribution (which may be caused by the content itself or block effects) and local high-frequency noise.

[0040] The energy percentage of the k-th frequency band is directly derived from the normalized frequency band energy distribution generated in the previous step. It represents the proportion of the k-th frequency band's energy in the total residual energy of the entire frame, and is a dimensionless value between 0 and 1. (The last part, "all frequency bands," appears to be a separate, unrelated statement and is left untranslated.) The sum of all these is always equal to 1. For example, in the previous step, by analyzing a video frame containing a city street scene, the normalized frequency band energy distribution vector obtained is: Therefore, in this calculation, The value is 0.65. The value is 0.18, and so on.

[0041] The frequency band number is a system parameter preset based on the coding block size and the accuracy requirements of the perceptual model. For an 8x8 transform block, there are 64 frequency coefficients. To achieve a balance between computational complexity and analysis accuracy, these 64 coefficients are usually divided into 8 to 16 frequency bands. In this embodiment, using the aforementioned division method based on the characteristics of the human visual system, the frequency coefficients are divided into 8 frequency bands. Therefore, the frequency band number is set to 8. .

[0042] Let be the uniform proportion of the k-th frequency band. It represents the energy distribution of a theoretically completely random, unstructured residual signal across all frequency bands. In this ideal case, the energy would be uniformly distributed across all frequency bands, thus each band would have the same proportion. Its calculation formula is: Due to the aforementioned set number of frequency bands Therefore, for all k from 1 to 8, The values ​​are all .

[0043] Let be the visual perception weight for the k-th frequency band. This weight vector is used to simulate the difference in human eye sensitivity to distortions at different spatial frequencies. Its value is set based on a large amount of psychophysical experimental data. Specifically, the acquisition process involves constructing a database containing hundreds of standard test images, adding quantized noise of a specific frequency band to each image, and having a group of observers subjectively rate the quality of the processed images (e.g., using a 5-point absolute rating system). After collecting all the rating data, for each frequency band, a functional relationship is established between the introduced noise energy and the decrease in subjective quality score. The slope of this functional relationship or a key parameter quantifies the human eye's sensitivity to noise in that frequency band. After normalizing the sensitivity values ​​for all frequency bands, the weight vector is obtained. A typical weight vector (corresponding to Z=8) is: The intermediate frequency has a higher weight, while the DC and mid-to-high frequency have lower weights. However, the weight of the highest frequency (k=8) is artificially increased to 2.0 to specifically punish high-frequency artifacts such as mosquito noise.

[0044] The balancing coefficient is used to adjust the relative importance of the KL divergence term and the high-frequency penalty term in the final spatial frequency spectrum divergence value. Its value is determined through parameter optimization on a video set containing various typical compression distortions (such as blockiness, ringing, and mosquito noise). Specifically, an evaluation index is set, for example, the calculated... The consistency between the value and human observers' judgments of the type and degree of video distortion (measured by classification accuracy or correlation coefficient) is then iterated within a preset range (e.g., 0.05 to 0.4) with a step size of 0.01. For each frame of the video set, calculate all possible values. The value is evaluated, and its consistency with subjective judgment is assessed; the value that maximizes the consistency index is selected. As the final configuration value, experimental analysis shows that when When the value is 0.15, the model has the strongest ability to distinguish between different types of distortion, therefore, we set it to... .

[0045] Calculations based on parameters: The number of frequency bands is obtained by using the parameters described above in the previous steps. Balance coefficient Energy percentage sequence even proportion Visual perception weights .

[0046] First, calculate the KL divergence term: ; Summing yields: ; Next, calculate the high-frequency penalty terms: : ; ; ; ; The maximum value is 0.03.

[0047] Final calculation : ; The results show that the spatial frequency spectrum divergence of the current frame is 0.37595, which is relatively small. This indicates that the energy distribution of the residual signal is not much different from that of the uniform noise, and there are no obvious high-frequency energy spikes. This usually corresponds to two situations: one is that the video source itself contains a lot of random noise, and the other is that the video coding quality is very high, most of the structured information has been effectively predicted, and the remaining residual is close to white noise.

[0048] The steps to obtain the adaptive loop filter configuration are as follows: Based on the spatial frequency spectrum divergence value, compare the spatial frequency spectrum divergence value with the artifact threshold and the noise threshold. If the spatial frequency spectrum divergence value exceeds the artifact threshold, select strong filtering and enable edge enhancement mode. If the spatial frequency spectrum divergence value is between the two thresholds, select medium filtering and enable smoothing mode. If the spatial frequency spectrum divergence value is lower than the noise threshold, turn off the filtering module and generate an adaptive loop filtering configuration.

[0049] Specifically, based on the spatial frequency spectrum divergence value, the calculated... The value (e.g., 0.37595) is compared sequentially with two preset thresholds: an artifact threshold and a noise threshold. First, a video quality database is needed, containing thousands of video clips generated by various encoders at different bitrates. Each clip is accompanied by a subjective quality score (MOS) given by multiple professional evaluators, along with detailed artifact type annotations (e.g., "significant blockiness," "ringing," "clean image," etc.). The spatial frequency spectrum divergence value is calculated for each video frame in the database. Then, all clips labeled "clean image" are compared sequentially. "The divergence values ​​of frames with a MOS score higher than 4.5 are statistically analyzed, and the 90th percentile of the distribution is taken as the noise threshold. For example, if 90% of high-quality frames have a divergence value lower than 0.5, then the noise threshold is set to 0.5. Next, the divergence values ​​of all frames marked as "significant blockiness" or "severe ringing" with a MOS score lower than 2.0 are statistically analyzed, and the 10th percentile of the distribution is taken as the artifact threshold. For example, if 90% of low-quality frames have a divergence value higher than 1.5, then the artifact threshold is set to 1.5. During operation, the first step is to determine..." If the artifact threshold of 1.5 is exceeded, it indicates severe structural distortion in the image that is easily perceptible to the human eye. In this case, a command is sent to the loop filtering module to select a strong filtering strategy, which involves setting the boundary strength parameters and quantization step size adjustment threshold of the deblocking filter more aggressively. Simultaneously, an edge enhancement mode is enabled. This mode smooths block boundaries while detecting and sharpening the true edges in the image. If the value is not greater than 1.5, then it is further determined whether it is below the noise threshold of 0.5. If it is below, it indicates that the residual signal is close to random noise, the image quality is high, and no filtering or only very weak filtering is needed. At this time, a command is sent to completely shut down the loop filtering module. A value between 0.5 and 1.5 indicates that the image may have slight distortion or contain a lot of complex textures. In this case, a medium-strength filter is selected and the smoothing mode is enabled to obtain an adaptive loop filter configuration.

[0050] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A narrow-bandwidth-based video monitoring data transmission storage method, characterized by, The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps:

2. The narrow bandwidth based video surveillance data transmission and storage method according to claim 1, characterized in that, The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps:

3. The narrow bandwidth based video surveillance data transmission and storage method of claim 1, wherein, The method comprises the following steps: The method comprises the following steps:

4. The narrow bandwidth based video surveillance data transmission and storage method of claim 1, wherein, The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: The method comprises the According to the GOP structure selection result of the video monitoring data stream, the group frame length, key frame interval and reference frame level are written in the encoding initialization stage of the video monitoring data stream, applied to subsequent frame grouping in time sequence and switching boundaries are established, parameter resetting and buffer queue refreshing are performed at the cross-segment switching place, and an adaptive image group structure is generated.

5. The narrow bandwidth based video surveillance data transmission and storage method of claim 1, wherein, The obtaining step of the frame-level code rate allocation scheme is: According to the adaptive image group structure, the frame sequence position, frame type and reference level in the image group are read, the frame quota is allocated according to the frame sequence position and the reference level in combination with the scene activity score and the total code rate budget, and the frame quota in the key frame interval is recorded, so as to generate the frame-level code rate allocation scheme.

6. The narrow bandwidth based video surveillance data transmission and storage method of claim 1, wherein, The obtaining step of the prediction residual spectrum data is: According to the frame-level code rate allocation scheme, the coding unit target bits are divided according to the frame quota, the reference pixel buffer is established for each coding unit, and the prediction signal is generated, the difference block between the original pixel and the prediction signal is calculated, the block domain frequency domain transformation is performed, and the prediction residual block coefficient matrix is generated after quantization according to the quantization step; According to the prediction residual block coefficient matrix, the coefficients with the same frequency index are merged in the order of blocks, the amplitude distribution and energy proportion of each frequency index are counted, the frequency band division table is constructed, and the amplitude statistics and energy statistics of each frequency band are output, so as to generate the prediction residual spectrum data.

7. The narrow bandwidth based video surveillance data transmission and storage method of claim 1, wherein, The obtaining step of the spatial frequency spectrum divergence value is: According to the prediction residual spectrum data, the coefficients are divided into multiple frequency bands according to the frequency index, the square amplitudes of all coefficients in each frequency band are summed and divided by the total number of coefficients in the frequency band, the average energy of each frequency band is calculated, the total energy of the full frequency band is counted, and the average energy of each frequency band is divided by the total energy of the full frequency band to obtain an energy proportion sequence, so as to generate the normalized frequency band energy distribution; According to the normalized frequency band energy distribution, the spatial frequency spectrum divergence value is calculated.

8. The narrow bandwidth based video surveillance data transmission and storage method of claim 1, wherein, The obtaining step of the adaptive loop filtering configuration is: According to the spatial frequency spectrum divergence value, the spatial frequency spectrum divergence value is compared with the artifact threshold value and the noise threshold value, if the spatial frequency spectrum divergence value exceeds the artifact threshold value, strong filtering is selected and the edge enhancement mode is enabled, if the spatial frequency spectrum divergence value is between the two threshold values, medium filtering is selected and the smoothing mode is enabled, and if the spatial frequency spectrum divergence value is lower than the noise threshold value, the filtering module is closed, so as to generate the adaptive loop filtering configuration.

Images