Search

CN-121982782-A - Privacy protection identification method and system

CN121982782ACN 121982782 ACN121982782 ACN 121982782ACN-121982782-A

Abstract

The invention belongs to the technical field of biological feature recognition, and discloses a privacy protection recognition method and a privacy protection recognition system, wherein a binocular camera is used for detecting a target distance; the method comprises the steps of executing AI elimination on targets outside a threshold according to a distance threshold for an alarm picture, analyzing target distances and executing AI elimination by a cloud server for a video stream, measuring distance of human body parts, completely imaging if any part enters a limit, detecting limb interaction, completely imaging all targets when interaction exists, identifying neighbor gate opening and closing states, automatically shrinking a sensitive scene to 1/4 distance by using a conventional scene as a limit with 1/2 distance, and executing cache backtracking and grading processing. The method solves the problems of fixed sensing distance, incomplete privacy elimination, video plot breakage, poor scene suitability and the like in the prior art, and achieves the effects of thoroughly eliminating target space-time traces, guaranteeing video continuity and dynamically adapting to sensitive scenes.

Inventors

  • ZHOU DI
  • WANG WEIJIE
  • XU GUOZHU
  • WU SHENGJI
  • YE JIANYUN

Assignees

  • 杭州电子科技大学
  • 杭州领芯微电子有限公司

Dates

Publication Date
20260505
Application Date
20260407

Claims (10)

  1. 1. A method of privacy preserving identification comprising the steps of: step 1, detecting the distance between a person target and a camera through a binocular camera; Step 2, marking distances for all personnel targets according to the alarm pictures, determining imaging rules of the image targets according to the relation between the distances and a preset distance threshold, and executing AI elimination processing on the personnel targets with the distances larger than the distance threshold; Step 3, aiming at the video stream, analyzing the distance of the personnel target in the video frame by a cloud server, determining an imaging rule according to the distance threshold, and executing AI elimination processing on the personnel target with the distance larger than the distance threshold; Step 4, optimizing privacy target imaging in the video stream, wherein the method comprises the steps of ranging a human body part, performing complete imaging on a personnel target if the distance of any part is smaller than the distance threshold value, detecting multi-person limb interaction behaviors, and performing complete imaging on all interaction targets if limb interaction exists; And 5, dynamically adjusting the limit based on scene recognition, wherein the dynamic limit comprises detecting the opening and closing state of a neighbor gate body, dynamically adjusting the distance threshold from a first threshold to a second threshold according to the gate body state, and executing caching, backtracking analysis and privacy processing on the adjusted video stream.
  2. 2. The method for recognizing privacy protection according to claim 1, wherein the step 1 specifically comprises: Acquiring a physical focal length, a base line distance and a pixel size of the binocular camera, and converting the physical focal length into a pixel focal length; and calculating parallax based on a binocular range formula, and determining the distance Z between the personnel target and the camera according to the parallax.
  3. 3. The method for recognizing privacy protection according to claim 1, wherein the step 2 specifically comprises: Step 2.1, detecting personnel in the alarm picture by using a target detection model, and endowing each personnel target with a unique ID; Step 2.2, calculating the distance Z between each personnel target and the camera based on the parallax value obtained in the step 1, and forming an (ID, Z) value pair; step 2.3, comparing the distance Z of each personnel target with a preset distance threshold value: if Z is less than or equal to the threshold value, the personnel target does not incorporate the picture processing sequence, and complete imaging is reserved; if Z > threshold, the personnel goal incorporates AI elimination processing sequence; step 2.4. Performing an elimination process on the personnel targets incorporating the AI elimination processing sequence.
  4. 4. A method of identifying privacy protection as defined in claim 3, wherein step 2.4 comprises the specific steps of: 2.4.1, precisely dividing a pixel region of a personnel target through an image segmentation model to generate a Mask; step 2.4.2, extracting features of different levels by using an encoder, and carrying out multi-scale feature fusion; And 2.4.3, performing iterative generation and restoration by using a generation countermeasure network model, wherein the iterative generation and restoration comprises the steps of using expansion convolution to expand the low-resolution content of a perceived mask region, analyzing the context relation of the whole graph through an attention mechanism, copying similar textures through context attention matching surrounding image blocks, and using countermeasure training to make the generation region indistinguishable from a real background.
  5. 5. The method for recognizing privacy protection according to claim 1, wherein the step 3 specifically comprises: step 3.1, uploading a video stream acquired by a camera to a cloud server through a switch; Step 3.2, the server detects personnel targets in the video frame, assigns a unique ID for each target, calculates the distance Z between each target and the camera by combining the parallax data of the binocular camera, and forms an (ID, Z) value pair; Step 3.3, comparing the distance Z of each target with a preset distance threshold value: if Z is less than or equal to the threshold value, considered as an object within the private space of the home side, reserving complete imaging; if Z > threshold, regard as the goal in the privacy space of the neighbor, enter AI and dispel the processing sequence; step 3.4, AI elimination processing is carried out on the target to be eliminated; And 3.5, outputting the processed video stream, and storing the video stream in a cloud or for playback and viewing by a user.
  6. 6. The method for recognizing privacy protection according to claim 1, wherein the step 4 specifically comprises: Step 4.1, the cloud server divides the uploaded video into video segments according to the existence of the person in the video of the video camera, and screens out the video segments with the person targets; step 4.2, identifying each part of the human body by using a human body part segmentation model for the personnel target, and respectively measuring the distance of each part; If the distance between any part of the personnel target is smaller than a preset distance threshold, the personnel target is completely imaged, and AI elimination processing is not executed; Step 4.3, when a plurality of personnel targets exist in the video, detecting whether limb interaction behaviors exist among the targets; If limb interaction exists, performing complete imaging on all personnel targets with the interaction, and not performing AI elimination processing; and 4.4, for the video segment in which personnel invade the side space, except for the personnel targets which are determined to be completely imaged in the step 4.2 and the step 4.3, the rest of the personnel targets which are positioned outside the side space are replaced by adopting a pixelation processing or mosaic mode.
  7. 7. The method for recognizing privacy protection according to claim 1, wherein the step 5 comprises: The intelligent door lock is triggered by scene detection and limit switching, namely, the intelligent door lock is used for identifying the opening and closing states of the neighbor door locks through image frames acquired by the binocular cameras and combining a target detection algorithm, and judging whether the neighbors are in a door opening action or a door opening state by extracting the contour change of the door body, the rotation track of the door shaft and the relative position characteristics of the door body and the door frame; Step 5.2, dynamic limit rule setting: The conventional scene is that a neighbor door body is closed, no door opening action exists, and a first threshold value is maintained, wherein the first threshold value is one half of the distance between a side door lock and the neighbor door lock; And in the sensitive scene, detecting the neighbor door opening action or the door opening, and automatically adjusting the distance threshold value to a second threshold value which is one quarter of the distance between the side door lock and the neighbor door lock.
  8. 8. The privacy preserving identification method of claim 7, wherein the step 5 further comprises: step 5.3, video buffering and grading processing mechanism: Automatically caching the video stream with the subsequent preset duration from the initial moment of detecting the neighbor door opening action, temporarily storing the cached data in a local storage module and synchronously uploading the cached data to a cloud server for backup; After the caching period is finished, the cloud server performs backtracking analysis on the cached video, and judges whether the neighbors and related targets cross the adjusted second threshold value or not: if the line is not crossed, AI elimination processing is carried out on the images of the neighbors in the video, the scene in the door and the related sensitive areas; If the multi-target interaction exists, the complete imaging is carried out on all interaction targets even if only a single target crosses the line; and 5.4, a limit resetting mechanism, namely automatically resetting the distance threshold to a first threshold value if the neighbor door body is not detected to be closed again or the neighbor door body is restored to be closed after the caching period is ended, and automatically prolonging the caching period once every preset time interval until the neighbor door body is closed and resetting the limit if the neighbor is continuously in the door-opening state.
  9. 9. An identification system for privacy protection, characterized in that it is configured to implement the method of any one of claims 1-8, and comprises a front-end perception layer, a local computing layer, a cloud processing layer, and a user configuration module; The front end perception layer is deployed locally at the intelligent door lock, and comprises: the binocular camera module is used for acquiring image data; the local cache module is used for temporarily storing the video stream in the sensitive scene; The communication module is used for uploading the image data and the video stream to the cloud; The local computing layer is deployed locally on the intelligent door lock and comprises: The binocular distance measuring module is used for calculating the distance between the personnel target and the camera based on binocular vision; the scene recognition module is used for recognizing the switching state of the neighbor gate; The dynamic limit module is used for dynamically adjusting the distance threshold according to the door body state; the picture processing module is used for executing AI elimination processing based on a distance threshold value on personnel targets in the alarm picture; the cloud processing layer is deployed on a cloud server and comprises: The video stream processing module is used for executing target detection, distance calculation and AI elimination processing on the uploaded video stream; The limb interaction recognition module is used for detecting the limb interaction behaviors of multiple people and controlling the imaging rule; the video backtracking module is used for carrying out backtracking analysis and grading treatment on the cache video under the sensitive scene; The cloud storage module is used for storing the processed video stream; the user configuration module is used for configuring privacy limit, threshold definition and system management parameters for a user.
  10. 10. The privacy preserving identification system of claim 9 wherein the dynamic limit module comprises: the door body state detection unit is used for identifying the switch state of the neighbor door body through a target detection algorithm; the threshold dynamic adjustment unit is used for switching the distance threshold between the first threshold and the second threshold according to the door body state; And the dynamic limit generating unit is used for transmitting the adjusted threshold value to the picture processing module and the video stream processing module for execution.

Description

Privacy protection identification method and system Technical Field The invention belongs to the technical field of biological feature recognition, and particularly relates to a recognition method and a recognition system for privacy protection. Background The intelligent door lock supports one or more unlocking modes such as passwords, fingerprints, faces, finger veins and palm veins, and the intelligent door lock is usually matched with a camera with functions such as 24-hour shooting, face recognition and stranger alarming while the convenience of entering and exiting is improved, so that omnibearing security monitoring is realized. However, in the process of monitoring the security of the home gate and assisting in multi-biometric identification and verification, the camera can continuously capture the dynamics of the neighbors, inevitably records the access condition of the neighbors and the staff of the same person, and even can correlate the biometric related information of the neighbors through the functions of face recognition, infrared and the like. The scheme of the prior patent application number 202510069328.7 is that the intelligent door lock is controlled to acquire current infrared data of the infrared data acquisition equipment in response to the opening operation of a privacy marking function in the intelligent door lock by a user, the current infrared data is subjected to tag identification to obtain a tag identification result, a target privacy area is generated according to the tag identification result, and the target privacy area is stored, so that the infrared data acquired by the infrared data acquisition equipment is subjected to privacy processing according to the target privacy area when the intelligent door lock is used later. The current scheme mainly has two defects that 1, an infrared acquisition device is used in the scheme, but the sensing distance of a specific infrared device cannot be adapted to intelligent lock monitoring scenes with different distances, 2, any sensitive position can be shielded according to a recorded target privacy zone, and sensitive information such as the time and the number of the target appearance can be still left. Disclosure of Invention The invention aims to provide a privacy protection identification method and system so as to solve the technical problems. According to the invention, monitoring scenes with different space sizes are automatically adapted through a camera ranging technology, and personnel targets beyond the personal safety guard range are eliminated according to the distance imaging rules set by manufacturers or users, so that the problem that the personal privacy of the cameras is possibly leaked by neighbors is comprehensively solved. The specific technical scheme is as follows: a method of privacy preserving identification, comprising the steps of: The method comprises the steps of 1, detecting the distance between a personnel target and a camera through a binocular camera, 2, marking the distance for each personnel target aiming at an alarm picture, determining an imaging rule of an image target according to the relation between the distance and a preset distance threshold, executing AI elimination processing on the personnel target with the distance larger than the distance threshold, 3, analyzing the distance of the personnel target in a video frame aiming at a video stream by a cloud server, executing AI elimination processing on the personnel target with the distance larger than the distance threshold according to the distance threshold, 4, optimizing the imaging of a privacy target in the video stream, including ranging a human body part, executing complete imaging on the personnel target if the distance between any part is smaller than the distance threshold, detecting the interaction behavior of multiple personnel, executing complete imaging on all interaction targets if the interaction behavior of the limbs exists, and 5, executing dynamic limit adjustment based on scene recognition, including detecting the switching state of a neighbor gate body, dynamically adjusting the distance threshold from a first threshold to a second threshold according to the gate body state, executing cache and executing privacy analysis on the adjusted video stream. Further, the step 1 specifically includes: Acquiring a physical focal length, a base line distance and a pixel size of the binocular camera, and converting the physical focal length into a pixel focal length; calculating parallax based on a binocular range formula, and determining the distance Z between a personnel target and a camera according to the parallax; further, the step 2 specifically includes: Step 2.1, detecting personnel in the alarm picture by using a target detection model, and endowing each personnel target with a unique ID; Step 2.2, calculating the distance Z between each personnel target and the camera based on the parallax value obtained in the step 1, and