Search

CN-121999417-A - Intelligent data scheduling method and system for security protection of intelligent park

CN121999417ACN 121999417 ACN121999417 ACN 121999417ACN-121999417-A

Abstract

The invention relates to the technical field of image processing, in particular to an intelligent data scheduling method and system for intelligent park security, comprising the following steps: performing security monitoring on a target monitoring area by using a plurality of security monitoring devices to obtain a plurality of security videos, identifying a plurality of marked target image sets, confirming an abnormal target image set based on the plurality of marked target image sets and a face database, constructing a reference feature vector, acquiring a plurality of key frame sets, acquiring a plurality of target key frames by using the plurality of key frame sets and the reference feature vector, acquiring a plurality of security video fragments based on the plurality of target key frames, performing time sequence analysis on the plurality of security video fragments to obtain a behavior label, associating the behavior label with the abnormal target image set to obtain a security data label, and realizing intelligent data scheduling. The intelligent park security data scheduling method and system can improve efficiency and accuracy of intelligent park security data scheduling.

Inventors

  • HE FAN

Assignees

  • 厦门国瑞云卓科技有限公司

Dates

Publication Date
20260508
Application Date
20260226

Claims (10)

  1. 1. An intelligent data scheduling method for intelligent park security, which is characterized by comprising the following steps: Receiving an intelligent data scheduling instruction, and confirming an intelligent data scheduling environment based on the intelligent data scheduling instruction, wherein the intelligent data scheduling environment comprises a target monitoring area and a plurality of security monitoring devices; performing security monitoring on a target monitoring area by utilizing a plurality of security monitoring devices in an intelligent data scheduling environment to obtain a plurality of security videos, wherein the security videos correspond to the security monitoring devices one by one; Identifying a plurality of marked target image sets based on a plurality of security videos; confirming an abnormal target image set based on a plurality of marked target image sets and a pre-constructed face database, wherein the abnormal target image set comprises a plurality of abnormal target images; Constructing a reference feature vector based on a plurality of abnormal target images in the abnormal target image set; Acquiring a plurality of keyframe sets based on the abnormal target image set; Acquiring a plurality of target key frames by utilizing a plurality of key frame sets and reference feature vectors; acquiring a plurality of security video clips based on a plurality of target key frames; Performing time sequence analysis on the plurality of security video clips to obtain behavior tags; and associating the behavior tag with the abnormal target image set to obtain a security data tag, and realizing intelligent data scheduling based on the security data tag.
  2. 2. The intelligent data scheduling method for intelligent campus security of claim 1, wherein the identifying a plurality of marked target image sets based on a plurality of security videos comprises: Acquiring a plurality of non-target image frames by using a plurality of security monitoring devices, wherein the non-target image frames correspond to the security monitoring devices one by one; Constructing a park target recognition model by utilizing a plurality of non-target image frames; executing the following operations on each security video of the plurality of security videos: acquiring a video frame sequence based on a security video, wherein the video frame sequence comprises a plurality of video frames; obtaining a plurality of candidate identification target images by utilizing a plurality of video frames in a video frame sequence and a park target identification model; And performing feature screening on the candidate identification target images to obtain a plurality of marked target image sets.
  3. 3. The intelligent data scheduling method for intelligent park security as set forth in claim 2, wherein the feature screening the plurality of candidate recognition target images to obtain a plurality of marked target image sets includes: The following operation is performed for each of the plurality of candidate recognition target images: extracting core features of the candidate identification target images to obtain a target feature set, wherein the target feature set comprises target contour features and target size features; summarizing the target feature sets to obtain a plurality of target feature sets; acquiring a plurality of preliminary identification target images by utilizing a plurality of target feature sets and a pre-constructed invalid target feature threshold set; a plurality of sets of tagged target images are acquired using the plurality of preliminary identification target images.
  4. 4. The intelligent data scheduling method for intelligent campus security as claimed in claim 3, wherein the acquiring a plurality of sets of mark target images using a plurality of preliminary identification target images comprises: The following operation is performed for each of the plurality of preliminary identification target images: Graying is carried out on the preliminary identification target image, so that a graying identification target image is obtained; Acquiring a plurality of rectangular area images by using the graying identification target image; the following is performed for each of the plurality of rectangular area images: acquiring four vertex gray scale integral values of a rectangular area image, wherein the four vertex gray scale integral values comprise a first integral value, a second integral value, a third integral value and a fourth integral value; And calculating a rectangular gray scale characteristic by using the four vertex gray scale integral values, wherein the calculation formula of the rectangular gray scale characteristic is as follows: ; Wherein, the A rectangular gray scale feature is represented and, A fourth integrated value among the four vertex gray values is represented, The third integrated value among the four vertex gray values is represented, Represents the second integrated value among the four vertex gray values, Representing a first one of the four vertex integration values; acquiring a face feature recognition result by utilizing a pre-constructed face classifier and rectangular gray features, wherein the face feature recognition result comprises a non-face area and a face area; Taking a rectangular area image with the face characteristic recognition result being a face area as a face area local image; summarizing the partial images of the face areas to obtain a plurality of partial images of the face areas; Image fusion is carried out on the partial images of the plurality of face areas, so that a complete face image is obtained; performing face correction on the complete face image to obtain a corrected face image; summarizing the corrected face images to obtain a plurality of corrected face images; And carrying out cluster fusion on the plurality of corrected face images to obtain a plurality of marked target image sets.
  5. 5. The intelligent data scheduling method for intelligent park security as set forth in claim 4, wherein said performing face correction on the complete face image to obtain a corrected face image comprises: Acquiring 64 divided pixel blocks based on the whole face image, wherein the divided pixel blocks comprise a plurality of divided pixels; The following is performed for each of the 64 divided pixel blocks: the following is performed for each divided pixel in the divided pixel block: taking the divided pixel as a center pixel; acquiring 8 neighborhood pixels based on the center pixel; Performing binarization feature calculation by using the central pixel and 8 neighborhood pixels to obtain a face texture feature value; Summarizing the human face texture characteristic values to obtain a human face texture characteristic value set; Constructing a face gray level histogram by using the face texture characteristic value set; summarizing the face gray level histogram to obtain 64 face gray level histograms, wherein the face gray level histograms are in one-to-one correspondence with the divided pixel blocks; and carrying out angle correction on the whole face image by using the 64 face gray level histogram and the pre-constructed face template to obtain a corrected face image.
  6. 6. The intelligent data scheduling method for intelligent campus security of claim 5, wherein the constructing a reference feature vector based on a plurality of anomaly target images in the anomaly target image set comprises: the following operation is performed for each of the plurality of abnormal target images: Preprocessing an abnormal target image to obtain a preprocessed image; extracting a face depth feature vector by utilizing a pre-constructed face recognition model and a pre-processed image; constructing a gray level co-occurrence matrix by utilizing the preprocessed image; Acquiring a clothing feature vector by using the gray level co-occurrence matrix, wherein the clothing feature vector is a texture energy value; Performing body posture recognition on the preprocessed image to obtain a body type feature vector; summarizing the face depth feature vector, the clothing feature vector and the body type feature vector respectively to obtain a plurality of face depth feature vectors, a plurality of clothing feature vectors and a plurality of body type feature vectors; Normalizing each face depth feature vector of the face depth feature vectors, each clothing feature vector of the clothing feature vectors and each body type feature vector of the body type feature vectors to obtain normalized face depth feature vectors, normalized clothing feature vectors and normalized body type feature vectors; respectively splicing the plurality of normalized face depth feature vectors, the plurality of normalized clothing feature vectors and the plurality of normalized body type feature vector reference feature vectors to obtain a plurality of fusion feature vectors; And acquiring a reference feature vector by using the plurality of fusion feature vectors.
  7. 7. The intelligent data scheduling method for intelligent campus security of claim 6, wherein the acquiring a plurality of keyframe sets based on the anomaly target image set comprises: The following operations are performed for each of the abnormal target images in the abnormal target image set: confirming a security video source in a plurality of security videos by using the abnormal target image; acquiring a target time stamp of the abnormal target image; Intercepting a target security video clip in a security video source based on a target timestamp and a pre-confirmed time window; dividing the target security video segment by utilizing a frame division interval to obtain a key frame set; And summarizing the keyframe sets to obtain a plurality of keyframe sets.
  8. 8. The intelligent data scheduling method for intelligent campus security of claim 7, wherein the obtaining a plurality of target key frames using a plurality of key frame sets and reference feature vectors comprises: acquiring a plurality of image feature vectors of a plurality of key frames in a plurality of key frame sets, wherein the image feature vectors are in one-to-one correspondence with the key frames; the following is performed for each of a plurality of image feature vectors: Calculating feature similarity by using the image feature vector and the reference feature vector; Summarizing the feature similarities to obtain a plurality of feature similarities; comparing the multiple feature similarities with a preset target similarity threshold; taking the key frame with the feature similarity larger than or equal to the similarity threshold value as a target key frame; and summarizing the target key frames to obtain a plurality of target key frames.
  9. 9. The intelligent data scheduling method for intelligent campus security of claim 8, wherein the performing time-series analysis on the plurality of security video clips to obtain the behavior tag comprises: performing equipment classification on the plurality of security video clips to obtain a plurality of classified video clip sets, wherein the classified video clip sets comprise one or more classified video clips; the following is performed for each of a plurality of categorized video clip sets: segment fusion is carried out on a plurality of classified video segments in the classified video segment set, and a fused video is obtained; Summarizing the fused videos to obtain a plurality of fused videos; And performing behavior recognition on the multiple fused videos to obtain behavior tags.
  10. 10. An intelligent data scheduling system for intelligent campus security, the system comprising: The security video acquisition module is used for receiving the intelligent data scheduling instruction and confirming an intelligent data scheduling environment based on the intelligent data scheduling instruction, wherein the intelligent data scheduling environment comprises a target monitoring area and a plurality of security monitoring devices; performing security monitoring on a target monitoring area by utilizing a plurality of security monitoring devices in an intelligent data scheduling environment to obtain a plurality of security videos, wherein the security videos correspond to the security monitoring devices one by one; the abnormal target identification module is used for identifying a plurality of marked target image sets based on a plurality of security videos; confirming an abnormal target image set based on a plurality of marked target image sets and a pre-constructed face database, wherein the abnormal target image set comprises a plurality of abnormal target images; The abnormal behavior marking module is used for constructing a reference feature vector based on a plurality of abnormal target images in the abnormal target image set; Acquiring a plurality of keyframe sets based on the abnormal target image set; Acquiring a plurality of target key frames by utilizing a plurality of key frame sets and reference feature vectors; acquiring a plurality of security video clips based on a plurality of target key frames; Performing time sequence analysis on the plurality of security video clips to obtain behavior tags; And the scheduling data association module is used for associating the behavior tag with the abnormal target image set to obtain a security data tag, and intelligent data scheduling is realized based on the security data tag.

Description

Intelligent data scheduling method and system for security protection of intelligent park Technical Field The invention relates to the technical field of image processing, in particular to an intelligent data scheduling method and system for intelligent park security. Background Along with the development of the image processing field, the construction of the intelligent park has become the core direction of the digitalization of park management, wherein the security system is used as the core infrastructure of the intelligent park, directly related to the safety of personnel, equipment and property in the park, and is the key for guaranteeing the orderly operation of the park. Currently, security monitoring equipment is widely deployed in a security system of an intelligent park, after an intelligent data scheduling environment is confirmed by receiving an intelligent data scheduling instruction, video is collected, an abnormal target is identified and key frames and video fragments are extracted through the multi-security monitoring equipment, security data labels are generated through behavior analysis, and finally intelligent data scheduling is realized based on the labels. At present, security data is mainly scheduled through modes of total staff acquisition, unified transmission and centralized processing, and the scheduling purpose can be basically achieved by adopting a traditional method, but the problems of large redundancy of data, low recognition efficiency of abnormal targets and insufficient recognition accuracy exist. Therefore, optimizing the intelligent data scheduling method has important significance for improving the efficiency and the accuracy of intelligent park security data scheduling. Disclosure of Invention The invention provides an intelligent data scheduling method and a computer readable storage medium for intelligent park security, which mainly aim to improve the efficiency and the accuracy of intelligent park security data scheduling. In order to achieve the above purpose, the invention provides an intelligent data scheduling method for intelligent park security, comprising the following steps: Receiving an intelligent data scheduling instruction, and confirming an intelligent data scheduling environment based on the intelligent data scheduling instruction, wherein the intelligent data scheduling environment comprises a target monitoring area and a plurality of security monitoring devices; performing security monitoring on a target monitoring area by utilizing a plurality of security monitoring devices in an intelligent data scheduling environment to obtain a plurality of security videos, wherein the security videos correspond to the security monitoring devices one by one; Identifying a plurality of marked target image sets based on a plurality of security videos; confirming an abnormal target image set based on a plurality of marked target image sets and a pre-constructed face database, wherein the abnormal target image set comprises a plurality of abnormal target images; Constructing a reference feature vector based on a plurality of abnormal target images in the abnormal target image set; Acquiring a plurality of keyframe sets based on the abnormal target image set; Acquiring a plurality of target key frames by utilizing a plurality of key frame sets and reference feature vectors; acquiring a plurality of security video clips based on a plurality of target key frames; Performing time sequence analysis on the plurality of security video clips to obtain behavior tags; and associating the behavior tag with the abnormal target image set to obtain a security data tag, and realizing intelligent data scheduling based on the security data tag. Optionally, the identifying a plurality of marked target image sets based on a plurality of security videos includes: Acquiring a plurality of non-target image frames by using a plurality of security monitoring devices, wherein the non-target image frames correspond to the security monitoring devices one by one; Constructing a park target recognition model by utilizing a plurality of non-target image frames; executing the following operations on each security video of the plurality of security videos: acquiring a video frame sequence based on a security video, wherein the video frame sequence comprises a plurality of video frames; obtaining a plurality of candidate identification target images by utilizing a plurality of video frames in a video frame sequence and a park target identification model; And performing feature screening on the candidate identification target images to obtain a plurality of marked target image sets. Optionally, the feature screening is performed on the multiple candidate identification target images to obtain multiple marked target image sets, including: The following operation is performed for each of the plurality of candidate recognition target images: extracting core features of the candidate identification targ