Search

CN-121982656-A - Intelligent monitoring image recognition processing system for community emergency treatment

CN121982656ACN 121982656 ACN121982656 ACN 121982656ACN-121982656-A

Abstract

The invention relates to the technical field of image recognition, and discloses an intelligent monitoring image recognition processing system for community emergency treatment, which calculates the equivalent pixel size of a macro block by analyzing the resolution scaling of a video stream, and combining the horizontal line coordinates determined by perspective positioning and the texture periodicity, and measuring and calculating the artifact interference height to generate a suppression weighting band. Interference energy is then extracted from the background variance distribution and used as an attention penalty in the collaborative calculation of human behavior and variance sequences, thereby outputting treatment grade. The method solves the technical problem of confusion between coding noise and true anomaly, maintains extremely high identification stability in a low-code-rate environment, and is suitable for an intelligent community security management system.

Inventors

  • YOU LINING
  • ZHANG QINHUI

Assignees

  • 陕西奥润激光技术有限公司
  • 深圳鼎欣智能科技有限公司

Dates

Publication Date
20260505
Application Date
20260407

Claims (8)

  1. 1. An intelligent monitoring image recognition processing system for community emergency treatment, which is characterized by comprising: the video analysis module is used for acquiring the resolution scaling of the monitoring video stream and calculating the equivalent pixel size of the macro block under the reference according to the resolution scaling; the difference detection module is used for comparing the continuous picture frames of the monitoring video stream and calculating an image background difference map of the adjacent picture frames; the perspective analysis module is used for extracting the image vanishing points of the continuous picture frames to determine the longitudinal coordinates of the horizontal lines and calculating the texture cycle rate of the picture texture cycle along with the change of the longitudinal coordinates; The interference positioning module is used for calculating an artifact interference height based on the horizontal line longitudinal coordinates, the texture periodic rate and the resolution scaling, and generating a suppression weighting band taking the artifact interference height as a center; The energy extraction module is used for extracting artifact interference energy from the frequency domain distribution of the image background difference map in the suppression weighted band by referring to the macro block boundary basic frequency corresponding to the macro block equivalent pixel size; the state evaluation module is used for carrying out cooperative calculation on the input human body behavior sequence and a background picture difference sequence formed by the image background difference image, taking the artifact interference energy as an attention penalty term aiming at the background picture difference sequence, and outputting a community emergency response treatment grade.
  2. 2. The intelligent monitoring image recognition processing system for community emergency treatment according to claim 1, wherein obtaining a resolution scaling of a monitoring video stream and calculating a macroblock equivalent pixel size under a reference according to the resolution scaling comprises: Analyzing the monitoring video stream, extracting the original picture width and the original picture height of the current picture frame, and acquiring the reference picture width and the reference picture height preset by the system; The resolution scale is obtained by the following calculation formula: ; Wherein, the Representing the scale of the resolution in question, Representing the width of the original picture in question, Representing the height of the original picture in question, Representing the width of the reference picture in question, Representing the height of the reference picture, Representing a minimum function; obtaining a macro block reference pixel size preset by a video coding standard, and calculating the macro block equivalent pixel size through the following calculation formula: ; Wherein, the Representing the macroblock equivalent pixel size, Representing the macroblock reference pixel size.
  3. 3. The intelligent monitoring image recognition processing system for community emergency treatment according to claim 2, wherein the intelligent monitoring image recognition processing system is characterized in that image key points of the current picture frame and image key points of a previous picture frame are extracted for matching calculation, an affine transformation matrix is obtained, and the affine transformation matrix is utilized for carrying out space position alignment on the current picture frame to obtain a corrected picture frame; Carrying out logarithmic mapping calculation on the corrected picture frame to obtain the brightness average value of the whole frame; Extracting a preset first frame brightness reference mean value, and carrying out mean alignment calculation on the whole frame brightness mean value and the first frame brightness reference mean value to obtain a brightness normalization picture frame; And carrying out smoothing filtering treatment on the brightness normalization picture frame to obtain a current preprocessing picture frame, and executing the same treatment flow on the previous picture frame to obtain the previous preprocessing picture frame, wherein the preprocessing calculation formula is as follows: ; ; Wherein, the Representing the frame of the corrective picture, Indicating that the anti-zero offset constant is present, Representing a logarithmic mapping function, Representing the luminance average value of the whole frame, Representing the first frame luminance reference mean, Representing the brightness normalization picture frame, Representing the smoothing filter processing function, Representing the current pre-processed picture frame; Calculating an absolute difference map of the current preprocessed picture frame and the previous preprocessed picture frame, and taking the absolute difference map as the image background difference map, wherein the calculation formula is as follows: ; Wherein, the Representing a background disparity map of the image, Representing the coordinates of the picture pixels, Representing the brightness value corresponding to the current preprocessed picture frame, Representing the corresponding luminance value of the previous pre-processed picture frame, Representing an absolute difference function.
  4. 4. A system for intelligent monitoring image recognition processing of community emergency treatment according to claim 3, wherein extracting the image vanishing points of the continuous picture frames to determine horizontal line longitudinal coordinates and calculating a texture cycle rate of the picture texture cycle as a function of the longitudinal coordinates comprises: extracting a line segment set of the continuous picture frames in a set region of interest; Screening the line segment set according to a preset angle range, and obtaining the image vanishing point through random sampling consistency calculation; and extracting the vertical axis coordinate data of the image vanishing point, and carrying out picture boundary truncation processing on the vertical axis coordinate data to obtain the horizontal line vertical coordinate, wherein the calculation formula is as follows: ; Wherein, the Representing the horizontal line longitudinal coordinates, Vertical axis coordinate data representing the vanishing point of the image, Representing the height of the reference picture, The representation takes the function of the minimum value, Representing a maximum function; dividing the region of interest into a plurality of local image line bands according to a preset line span, and calculating an intra-line average brightness signal of each local image line band; Performing one-dimensional discrete Fourier transform on the average brightness signal in the line to obtain a frequency spectrum sequence, determining a main spectrum peak in a preset frequency searching range, screening out a target image line band meeting a preset significance condition, and calculating to obtain a picture texture period based on a frequency position corresponding to the main spectrum peak in the target image line band, wherein the calculation formula is as follows: ; Wherein, the Representing the period of the texture of the picture, Representing the width of the reference picture in question, Representing the frequency position corresponding to the main peak of the frequency spectrum; and performing linear fitting on the picture texture period of the target image line band and the corresponding line height difference value by using a weighted least square method, and calculating to obtain the texture period rate, wherein the calculation formula is as follows: ; Wherein, the Representing the period rate of the texture in question, Representing a set of target image line band coordinates involved in the fit, Representing the current row-wise coordinate of the current row, Representing the fitted weighting coefficients of the corresponding target image line band, Representing a summation computation function.
  5. 5. The intelligent monitoring image recognition processing system for community emergency treatment according to claim 4, wherein the macro block reference pixel size is obtained, the artifact interference height is calculated according to the following calculation formula: ; Wherein, the Representing the height of the artifact interference, Representing the horizontal line longitudinal coordinates, Representing the macroblock reference pixel size, Representing the period rate of the texture in question, Representing the resolution scale.
  6. 6. The intelligent monitoring image recognition processing system for community emergency treatment of claim 5, wherein calculating an artifact interference height based on the horizontal line longitudinal coordinates, the texture periodicity and the resolution scaling and generating a suppression weighting band centered on the artifact interference height comprises: the reference picture height is obtained, the artifact interference height is limited in numerical range, and a limiting formula is as follows: ; Wherein, the Representing the height of the reference picture, The representation takes the function of the minimum value, Representing a maximum function; Multiplying the reference picture height by a preset proportionality coefficient to obtain a bandwidth parameter, traversing to obtain a current line coordinate in a picture, and calculating natural index operation data of the current line coordinate and the artifact interference height as initial weighted band data, wherein a calculation formula is as follows: ; Wherein, the Representing the initial weighted band data of the set, Representing the current row-wise coordinate of the current row, Which represents the bandwidth parameter in question, Representing a natural exponent arithmetic function; normalizing the initial weighted band data along the longitudinal direction of the picture to obtain normalized data as the suppression weighted band, wherein the processing formula is as follows: ; Wherein, the Representing the suppression weight band in question, Representing the normalized accumulated row index, Representing a summation computation function.
  7. 7. The intelligent monitoring image recognition processing system for community emergency treatment according to claim 6, wherein extracting artifact interference energy from the frequency domain distribution of the image background disparity map with reference to the macroblock boundary fundamental frequency corresponding to the macroblock equivalent pixel size in the suppression-weighted band comprises: calculating the pixel difference value in the horizontal direction of the current preprocessed picture frame to obtain an image horizontal gradient amplitude value, wherein the calculation formula is as follows: ; Wherein, the Representing the image horizontal gradient magnitude, Representing a current pixel brightness value corresponding to the current pre-processed picture frame, Representing the brightness value of the previous pixel in the same row corresponding to the current preprocessed picture frame, Representing an absolute difference function; performing line-wise Fourier transform on the horizontal gradient amplitude of the image to obtain a gradient spectrum sequence; dividing the product of the reference picture width and the resolution scaling by the macroblock reference pixel size and rounding down to obtain the macroblock boundary base frequency, with the following calculation formula: ; Wherein, the Representing the macroblock boundary base frequency, Representing the width of the reference picture in question, Representing the scale of the resolution in question, Representing the macroblock reference pixel size, Representing a downward rounding function; Calculating a harmonic set based on the macroblock boundary fundamental frequency, extracting the spectrum amplitude of the gradient spectrum sequence under the harmonic set, and carrying out longitudinal weighted summation by utilizing the suppression weighted band to obtain the total amount of original interference energy, wherein the calculation formula is as follows: ; Wherein, the Representing the total amount of said original interference energy, Representing the height of the reference picture, The vertical coordinates of the picture are represented, Representing the suppression weight band in question, Representing the harmonic set The number of harmonics in (a) is, Representing the amplitude extraction function corresponding to the gradient spectrum sequence, Representing a summation calculation function; Calculating a full-image mean value of the image horizontal gradient amplitude as global gradient reference energy, normalizing the total amount of the original interference energy by using the global gradient reference energy to obtain the artifact interference energy, wherein the calculation formula is as follows: ; ; Wherein, the Representing the said artifact interference energy, Representing the global gradient reference energy, The lateral coordinates of the picture are represented, Indicating a preset zero-offset prevention constant of the system.
  8. 8. The intelligent monitoring image recognition processing system for community emergency treatment according to claim 7, wherein the collaborative calculation of the input human behavior sequence and the background picture difference sequence composed of the image background difference map is performed, the artifact interference energy is used as an attention penalty term for the background picture difference sequence, and the community emergency response treatment level is output, comprising: The human body behavior sequence and the background picture difference sequence are respectively input into a coding network to obtain a behavior characterization vector and a difference characterization vector, and the calculation formula is as follows: ; ; Wherein, the The representation of the behavior characterization vector is performed, Representing a network from which the behavior characterization vector is extracted, Representing a network from which the disparity characterization vector is extracted, Representing the sequence of human behavior in question, The difference token vector is represented as such, Representing the background picture difference sequence; Calculating a first attention score for the behavior characterization vector and a second attention score for the difference characterization vector, punishing the second attention score by using the artifact interference energy, and acquiring a collaborative fusion characterization vector according to the punished second attention score and the first attention score, wherein the calculation formula is as follows: ; ; ; ; Wherein, the The first attention score is represented as such, Representing a transpose of the behavior scoring weight vector, A second attention score before the penalty is indicated, Representing the transpose of the differential scoring weight vector, Representing a preset penalty coefficient, Representing the said artifact interference energy, Representing the normalized weight of the behavior, The normalized weights of the differences are represented, Representing the normalized exponential function of the sample, Representing the collaborative fusion token vector; calculating the norm of the collaborative fusion characterization vector as abnormal state intensity, and extracting state spreading trend parameters by combining a preset time window, wherein the calculation formula is as follows: ; ; ; Wherein, the Representing the intensity of the abnormal state, Representing a two-norm calculation function, Representing the current intensity of the smoothing signal, Representing the coefficient of smoothing and the coefficient of smoothing, Representing the smooth intensity at the previous moment, Representing the state-propagation trend parameter, Representing the smooth intensity before the preset interval frame, Representing the preset time window interval; and carrying out weighted summation on the abnormal state intensity and the state spreading trend parameter, outputting the community emergency response treatment grade according to a preset threshold, wherein the calculation formula is as follows: ; ; Wherein, the The overall rating value is represented by a combination of values, The trend weight coefficient is represented as a function of the trend, Representing the community emergency response disposition level, A first preset threshold is indicated and, Representing a second preset threshold.

Description

Intelligent monitoring image recognition processing system for community emergency treatment Technical Field The invention relates to the technical field of image recognition, in particular to an intelligent monitoring image recognition processing system for community emergency treatment. Background With the continuous improvement of intelligent community management systems, automatic research and judgment of emergency events by utilizing visual monitoring has become a mainstream means. The existing monitoring and identification scheme mainly relies on attribute identification of fluctuation and movement tracks of pixel sequences in a picture. However, in a real community defense area environment, there is a convergence path extending along a perspective line in a critical area such as a main road, a gallery, an entrance and a gateway, or there is a regular arrangement layout formed by fences, gates, building outer walls, and the like. In a sudden emergency state, as the concurrent access of multiple paths of videos is very easy to cause the rapid increase of network bandwidth load, a video transmission system generally triggers an adaptive adjustment mechanism, and the transmission consistency is ensured by reducing the resolution or the compression code rate. In this environment, pixel block boundary effects generated by video coding overlap with regular elements inherent in a picture in spatial arrangement, thereby inducing a large number of spurious fluctuation signals in the picture. The existing identification logic is difficult to effectively filter the interference signals caused by the code degradation in a low-image-quality environment, so that false alarms frequently occur in an emergency period when the system is in urgent need of accurate decision, and the reliability of community emergency response decisions is severely limited. Disclosure of Invention The invention provides an intelligent monitoring image recognition processing system for community emergency treatment, which solves the technical problems in the background technology. The invention provides an intelligent monitoring image recognition processing system for community emergency treatment, which comprises the following components: the video analysis module is used for acquiring the resolution scaling of the monitoring video stream and calculating the equivalent pixel size of the macro block under the reference according to the resolution scaling; the difference detection module is used for comparing the continuous picture frames of the monitoring video stream and calculating an image background difference map of the adjacent picture frames; the perspective analysis module is used for extracting the image vanishing points of the continuous picture frames to determine the longitudinal coordinates of the horizontal lines and calculating the texture cycle rate of the picture texture cycle along with the change of the longitudinal coordinates; The interference positioning module is used for calculating an artifact interference height based on the horizontal line longitudinal coordinates, the texture periodic rate and the resolution scaling, and generating a suppression weighting band taking the artifact interference height as a center; The energy extraction module is used for extracting artifact interference energy from the frequency domain distribution of the image background difference map in the suppression weighted band by referring to the macro block boundary basic frequency corresponding to the macro block equivalent pixel size; the state evaluation module is used for carrying out cooperative calculation on the input human body behavior sequence and a background picture difference sequence formed by the image background difference image, taking the artifact interference energy as an attention penalty term aiming at the background picture difference sequence, and outputting a community emergency response treatment grade. Aiming at the picture noise interference generated when the concurrent response causes the bandwidth limitation in the community monitoring, the invention realizes the precision of the positioning of the compression artifacts and the dynamic of the interference filtering by establishing the internal association between the spatial domain layout of the image and the coding mechanism, obviously reduces the false alarm rate caused by the degradation of the image quality, ensures that the system can still output a high-reliability disposal instruction under the extreme network condition, and greatly strengthens the emergency response efficiency and the resource allocation precision of the base layer management system. Drawings FIG. 1 is a workflow diagram of an intelligent monitoring image recognition processing system for community emergency treatment of the present invention; FIG. 2 is a graph of experimental data of the present invention. Detailed Description The subject matter described herein will now be discusse