Search

CN-122024162-A - Visual intelligence-based power station abnormal personnel intrusion recognition method and system

CN122024162ACN 122024162 ACN122024162 ACN 122024162ACN-122024162-A

Abstract

The invention relates to the technical field of power station safety monitoring, and discloses a method and a system for identifying abnormal personnel invasion of a power station based on visual intelligence, wherein the method comprises the following steps of carrying out self-adaptive equalization pretreatment on a monitoring image; extracting and optimizing a motion area boundary based on an optical flow field, acquiring a foreground target outline through density clustering, obtaining a target mask through morphological processing, extracting gradient histogram features of a target sub-image according to the foreground target outline, performing similarity matching with an authorized personnel feature library to generate a verification mark, and performing sensitive area overlapping and behavior track analysis on an unauthorized target according to a verification result to trigger an intrusion alarm. The method can stably extract the moving target and accurately identify the personnel identity under complex illumination and environment, obviously reduce false alarm and missed detection and improve the intellectualization and reliability of the security protection of the power station.

Inventors

  • ZHANG JIN
  • WU ZHANHENG
  • ZHAN DONGDONG
  • LIU HUANHUAN
  • WU XIAOYE

Assignees

  • 紫金龙净清洁能源有限公司
  • 西藏麻米紫金龙净清洁能源有限公司

Dates

Publication Date
20260512
Application Date
20260202

Claims (9)

  1. 1. The utility model provides a power station abnormal personnel invasion identification method based on visual intelligence which is characterized by comprising the following steps: after an original image sequence is obtained, gray level conversion is carried out, pixel brightness distribution is adjusted and detail characteristics are enhanced on the converted gray level image sequence, boundary information is extracted and optimized, and an equalized image is obtained after denoising treatment; calculating an optical flow vector field according to the equalized image, extracting pixel points with the amplitude exceeding a motion intensity threshold value, connecting discontinuous boundaries and tracking pixel displacement, and determining a motion area boundary after smoothing filtering; Extracting foreground pixel clusters from the boundary of the motion area, and judging a potential target area if the density of the foreground pixel clusters exceeds a preset density threshold value to obtain a target contour map; Generating a binary mask according to the target profile, removing isolated noise points in the binary mask to obtain a first mask, analyzing and combining a communication component, smoothing and repairing a broken part to determine a complete target mask; Extracting a target sub-image from the equalization image based on the complete target mask, calculating a gradient, counting to generate a gradient direction histogram, and obtaining a gradient histogram feature vector for identity verification after direction interval weighting and vector normalization processing; retrieving similar feature vectors from a pre-established authorized personnel feature library, and judging the authorized personnel if the similarity score is higher than a preset similarity threshold value to obtain a verification passing identifier; and filtering the potential target area according to the verification passing mark, and triggering an intrusion alarm mechanism if the potential target area does not pass the verification passing mark, so as to obtain a final intrusion detection result.
  2. 2. The visual intelligence-based power station anomaly personnel intrusion recognition method according to claim 1, wherein the steps of obtaining an original image sequence, performing gray level conversion, adjusting pixel brightness distribution and enhancing detail characteristics for the converted gray level image sequence, extracting and optimizing boundary information, and obtaining an equalized image after denoising, comprise: Acquiring an original image sequence containing a plurality of frames of continuous monitoring pictures from a monitoring camera; carrying out gray level conversion treatment on each frame of image in the original image sequence to obtain a gray level image sequence; Adopting self-adaptive histogram equalization to adjust the pixel brightness distribution of each frame in the gray image sequence to obtain a brightness equalization image sequence; Carrying out local contrast enhancement treatment on each frame of image in the brightness equalization image sequence, and enhancing detail characteristics to obtain a detail enhancement image sequence; Extracting and optimizing boundary information in the detail enhancement image sequence through edge detection to obtain a clear image sequence; And carrying out noise suppression on the clear image sequence by adopting median filtering to obtain a final equalized image.
  3. 3. The visual intelligence-based power station anomaly personnel intrusion recognition method according to claim 1, wherein calculating an optical flow vector field according to the equalized image, extracting pixel points with an amplitude exceeding a motion intensity threshold value, connecting discontinuous boundaries and tracking pixel displacement, and determining a motion area boundary after smoothing filter processing comprises: calculating displacement vectors of pixels between successive frames by adopting an optical flow algorithm based on the equalized image to obtain a dense optical flow vector field; determining a motion intensity threshold according to the amplitude distribution of the dense optical flow vector field, extracting a pixel point set with the amplitude exceeding the threshold, and forming a preliminary motion area mask; Removing isolated noise points and connecting intermittent boundaries by morphological closing operation on the preliminary motion region mask to obtain a boundary contour line of the continuous motion region; and smoothing the point set of the boundary contour line of the continuous motion area by adopting a Gaussian smoothing filter to generate a smooth boundary curve, and determining the boundary of the motion area in the monitoring picture.
  4. 4. The visual intelligence-based power station abnormal personnel intrusion recognition method according to claim 1, wherein the steps of extracting foreground pixel clusters from the boundary of the motion area, judging the foreground pixel clusters as potential target areas if the density of the foreground pixel clusters exceeds a preset density threshold value, and obtaining a target profile map comprise the following steps: Extracting a foreground pixel point coordinate set from an internal area defined by the boundary of the motion area, and dividing foreground pixel clusters by adopting a density clustering algorithm to obtain cluster center coordinates and cluster point number; if the number of the cluster points exceeds the preset upper limit of the density threshold, marking the cluster points as potential target areas, and determining a pixel point set of the potential target areas as a central area point set; Performing boundary expansion on the central region point set and connecting adjacent cluster boundaries to form an expansion region contour line; Overlapping the original contour line in the boundary of the motion area and the contour line of the expansion area, and generating a refining target contour map after removing overlapping parts; and carrying out pixel connected domain analysis on the refined target profile and filling the internal cavity to obtain a complete target profile.
  5. 5. The visual intelligence based power station anomaly personnel intrusion recognition method according to claim 4, wherein generating a binary mask according to the target profile, removing isolated noise points in the binary mask to obtain a first mask, analyzing and merging communication components, smoothing and repairing broken parts to determine a complete target mask, and comprising: Generating a binary mask according to the target profile, removing isolated noise points from the binary mask by adopting morphological corrosion operation to obtain a first mask which is primarily cleaned; Analyzing the communication component of the first mask, identifying and combining the pixel groups which are adjacent in space and meet the preset area condition, and determining a communication region set; performing boundary smoothing processing on the connected region set by applying expansion operation to obtain a smoothed region boundary; and superposing and comparing the smooth area boundary with the original contour line, repairing the broken part by adopting a boundary tracking algorithm, and determining a final complete target mask.
  6. 6. The visual intelligence-based power station abnormal personnel intrusion recognition method according to claim 1, wherein the steps of extracting target sub-images from the equalized images based on the complete target mask, calculating gradients and statistically generating gradient direction histograms, obtaining gradient histogram feature vectors for identity verification after direction interval weighting and vector normalization processing comprise the steps of: separating a pixel region corresponding to the complete target mask from the equalized image to obtain a target sub-image; calculating the gray gradient amplitude and direction of the target sub-image, and generating a gradient direction histogram through statistics; Dividing a direction interval according to the gradient direction histogram, and accumulating frequencies to generate a direction weighted histogram vector; And carrying out normalization processing on the direction weighted histogram vector to obtain a gradient histogram feature vector for identity verification.
  7. 7. The visual intelligence based power station anomaly personnel intrusion recognition method according to claim 6, wherein retrieving similar feature vectors from a pre-established authorized personnel feature library, and if the similarity score is higher than a pre-set similarity threshold, determining that the person is authorized, and obtaining a verification passing identifier comprises: Extracting a stored standardized feature vector from a pre-established authorized personnel feature library; Calculating the similarity between the gradient histogram feature vector and each standardized feature vector in the authorized personnel feature library to obtain a similarity score matrix; Selecting the highest value in the similarity score matrix, and if the highest value is higher than a preset similarity threshold value, judging that the matching is successful; and associating the identity information of the authorized personnel corresponding to the successfully matched feature vector with the moving target to be checked currently, and generating a check passing identifier.
  8. 8. The visual intelligence based power station anomaly personnel intrusion identification method of claim 7, wherein filtering potential target areas according to the verification through identification, and triggering an intrusion alarm mechanism if the potential target areas do not pass through the verification to obtain a final intrusion detection result, comprising: If the moving target does not obtain the verification passing identification, marking the potential target area as an unauthorized target area; Comparing the space positions of the unauthorized target area and a predefined intrusion sensitive area, and judging the unauthorized target area and the predefined intrusion sensitive area as suspicious intrusion areas if the unauthorized target area and the predefined intrusion sensitive area overlap with any intrusion sensitive area; if the suspicious intrusion area has an abnormal motion path and the speed or direction change exceeds a preset safety threshold, triggering an intrusion alarm mechanism to obtain a final intrusion detection result.
  9. 9. The utility model provides a power station abnormal personnel invasion identification system based on visual intelligence which characterized in that includes: the image preprocessing and equalizing module is used for carrying out gray level conversion after acquiring an original image sequence, adjusting pixel brightness distribution and strengthening detail characteristics on the converted gray level image sequence, extracting and optimizing boundary information, and obtaining an equalized image after denoising; The motion area detection and boundary determination module calculates an optical flow vector field according to the equalized image, extracts pixel points with the amplitude exceeding a motion intensity threshold value, connects the intermittent boundary and tracks the pixel displacement, and determines the boundary of the motion area after smoothing filter processing; The foreground target extraction and contour generation module extracts foreground pixel clusters from the boundary of the motion region, and judges a potential target region if the density of the foreground pixel clusters exceeds a preset density threshold value to obtain a target contour map; The target mask optimization module generates a binary mask according to the target profile graph, obtains a first mask after isolated noise points in the binary mask are removed, analyzes and merges a communication component, and determines a complete target mask after smoothing and repairing a broken part; the feature extraction module is used for extracting a target sub-image from the equalization image based on the complete target mask, calculating gradients and counting to generate a gradient direction histogram, and obtaining a gradient histogram feature vector for identity verification after direction interval weighting and vector normalization processing; the identity verification module is used for retrieving similar feature vectors from a pre-established authorized person feature library, judging the authorized person if the similarity score is higher than a preset similarity threshold value, and obtaining verification passing identification; and the intrusion alarm decision-making module filters the potential target area according to the verification passing mark, and if the potential target area does not pass the verification passing mark, an intrusion alarm mechanism is triggered, and a final intrusion detection result is obtained.

Description

Visual intelligence-based power station abnormal personnel intrusion recognition method and system Technical Field The invention relates to the technical field of power station safety monitoring, in particular to a method and a system for identifying abnormal personnel invasion of a power station based on visual intelligence. Background The field of power station safety monitoring directly relates to the stable operation and public safety of national energy infrastructure, and once abnormal personnel invasion occurs, equipment damage and power interruption even a larger range of safety risks can be caused, so that real-time accurate identification of invasion behaviors becomes a key link for guaranteeing the safety of the power station. In the prior art, a motion detection method based on background modeling is adopted in a common intrusion detection scheme in the industrial vision field, for example, a static background is established for a monitoring video through a Gaussian mixture model, a foreground moving target is extracted by a frame difference method, and whether the abnormal intrusion behavior is judged by combining contour analysis or track tracking. However, in the prior art, it is generally assumed that the background of the monitored scene is relatively stable, but in an outdoor complex environment such as a power station, factors such as illumination change, weather conversion, vegetation growth and the like can cause continuous change of the background, so that updating of the background model is delayed, and foreground extraction is unstable. The system is easy to misjudge shadows, reflection, vegetation shaking and the like as targets, causes a large number of misinformation, influences the accuracy of subsequent feature extraction and identity verification, and causes the risks of missed detection and misinformation. Therefore, the prior art has the problems of poor adaptability to complex environment changes, high false detection rate caused by easy illumination and weather interference. Disclosure of Invention The invention provides a visual intelligence-based power station abnormal personnel intrusion recognition method and system, which are used for solving the problems of poor adaptability to complex environment changes, easiness in illumination and high false detection rate caused by weather interference in the prior art. In order to solve the technical problems, the invention provides a method for identifying abnormal personnel invasion of a power station based on visual intelligence, which comprises the following steps: after an original image sequence is obtained, gray level conversion is carried out, pixel brightness distribution is adjusted and detail characteristics are enhanced on the converted gray level image sequence, boundary information is extracted and optimized, and an equalized image is obtained after denoising treatment; calculating an optical flow vector field according to the equalized image, extracting pixel points with the amplitude exceeding a motion intensity threshold value, connecting discontinuous boundaries and tracking pixel displacement, and determining a motion area boundary after smoothing filtering; Extracting foreground pixel clusters from the boundary of the motion area, and judging a potential target area if the density of the foreground pixel clusters exceeds a preset density threshold value to obtain a target contour map; Generating a binary mask according to the target profile, removing isolated noise points in the binary mask to obtain a first mask, analyzing and combining a communication component, smoothing and repairing a broken part to determine a complete target mask; Extracting a target sub-image from the equalization image based on the complete target mask, calculating a gradient, counting to generate a gradient direction histogram, and obtaining a gradient histogram feature vector for identity verification after direction interval weighting and vector normalization processing; retrieving similar feature vectors from a pre-established authorized personnel feature library, and judging the authorized personnel if the similarity score is higher than a preset similarity threshold value to obtain a verification passing identifier; and filtering the potential target area according to the verification passing mark, and triggering an intrusion alarm mechanism if the potential target area does not pass the verification passing mark, so as to obtain a final intrusion detection result. In a second aspect, the present invention provides a system for identifying abnormal personnel intrusion in a power station based on visual intelligence, comprising: the image preprocessing and equalizing module is used for carrying out gray level conversion after acquiring an original image sequence, adjusting pixel brightness distribution and strengthening detail characteristics on the converted gray level image sequence, extracting and optimizing boundary informati