CN-121985179-A - Deep learning and behavior recognition-based children anti-addiction control method and terminal

CN121985179ACN 121985179 ACN121985179 ACN 121985179ACN-121985179-A

Abstract

The invention discloses a children anti-addiction control method and a terminal based on deep learning and behavior recognition, which relate to the technical field of computer vision and comprise the following steps of analyzing a watching object in front of the terminal in real time; the method comprises the steps of identifying that a current watching object is a child identity, calculating the watching distance between the current watching object and a screen, which is the child identity, by constructing a three-dimensional model of the child face, carrying out real-time judgment and analysis on the watching distance based on the determined child identity and the watching distance, controlling to carry out primary early warning when the watching distance is continuously detected to be lower than a preset safety threshold value in preset time, continuously monitoring the watching behavior of the current watching object of the child identity, and carrying out grading intervention reminding and control processing according to a preset differential gradient adjustment algorithm when the watching behavior of the child is detected to violate a preset rule. The invention has high accuracy rate of identifying the identity of the viewer, strong environmental adaptability, and quick and accurate efficiency of intervention and processing of the behavior of the viewer.

Inventors

WANG ZHIGUO
HUANG FEN
NIE HAI
LIU HAI

Assignees

深圳市酷开网络科技股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260130

Claims (10)

1. The children anti-addiction control method based on deep learning and behavior recognition is characterized by comprising the following steps of: Real-time analysis is carried out on the watching object in front of the terminal through a preset improved network model, whether the current watching object is a child identity or not is identified, and the occasionally-appearing watching object is eliminated; When the current watching object is identified as the identity of the child, calculating the watching distance between the current watching object which is the identity of the child and the screen by constructing a three-dimensional model of the face of the child and adopting a dynamic calibration flow; Based on the determined identity and the viewing distance of the child, carrying out real-time judgment and analysis on the viewing distance, and controlling to carry out primary early warning when the continuous preset time detects that the viewing distance is continuously lower than a preset safety threshold value; when the watching behavior of the child is monitored to violate the preset rule, performing grading intervention reminding and control processing according to a preset differential gradient adjustment algorithm.
2. The children anti-addiction control method based on deep learning and behavior recognition according to claim 1, wherein the step of analyzing the viewing object before the terminal in real time through a preset improved network model, recognizing whether the current viewing object is a child identity, and excluding the viewing object which occurs occasionally comprises: The method comprises the steps of collecting a public data set and video images collected by a built-in camera of a terminal in advance, marking the video images, defining a child face area, defining interference objects of adult faces and pets, and constructing a multi-scene child face data set; Carrying out light improvement on the YOLOv model, replacing a standard convolution layer with a phantom network module, and optimizing a loss function; Analyzing the pictures of the watched objects obtained by the terminal camera in real time through the improved YOLOv model, identifying whether the current watched objects are child identities or not, and eliminating the watched objects which occur occasionally; When the current viewing object is detected as child identity for a continuous predetermined time, it is determined that the current viewing object is identified as child identity.
3. The children anti-addiction control method based on deep learning and behavior recognition of claim 1, wherein when recognizing that the current viewing object is a child identity, the step of calculating a viewing distance between the current viewing object and a screen, which is the child identity, by constructing a three-dimensional model of the child face and adopting a dynamic calibration flow comprises: Capturing a facial image of the current watching object of the child identity through a built-in camera, performing dynamic calibration, calculating the focal length of the camera and the physical size of a sensor, and establishing a mapping relation between a pixel coordinate system and a real space scale; Constructing a three-dimensional facial feature model, capturing facial images of a user in real time through a built-in camera of the terminal, extracting coordinate data of preset three-dimensional feature points, setting a dynamic calibration flow, selecting the inner angles of the nose tip, the left and right eyes and the center of the chin as key ranging reference points, and measuring and calculating the Euclidean distance between the current watching of the identity of the child and the screen in real time to be used as the watching distance.
4. The method for controlling children's anti-addiction based on deep learning and behavior recognition according to claim 1, wherein after the child identity and the viewing distance are determined, real-time judgment and analysis are performed on the viewing distance, when the viewing distance is continuously detected to be lower than a preset safety threshold for a predetermined time, the first-level early warning is controlled, and the step of continuously monitoring the viewing behavior of the current viewing object of the child identity comprises the following steps: A double-thread monitoring system is constructed to conduct real-time analysis on the dynamic behaviors of children, wherein a thread I acquires a built-in camera data stream in real time, calculates the viewing distance between eyes of a user and a screen, and writes the viewing distance data into an annular buffer area; when a sliding window early warning mechanism is entered for judgment, setting a sliding window early warning mechanism based on a time sequence, carrying out exponential decay weighted average calculation on the watching distance data by a window unit of a first preset time, and triggering a first-stage early warning when the weighted average value of continuous preset window units is continuously lower than a safety threshold value; The viewing behavior of the current viewing object of the child identity is continuously monitored.
5. The children anti-addiction control method based on deep learning and behavior recognition according to claim 1, wherein when the watching behavior of the children is monitored to violate a preset rule, the step of executing hierarchical intervention reminding and control processing according to a preset differential gradient adjustment algorithm comprises: When the accumulated watching exceeds the second preset time, the control is switched into a linear attenuation mode, the threshold value is gradually tightened, and finally, the strict threshold value is reached at the third preset time; Controlling a preset differential gradient adjustment algorithm to realize graded intervention, and when the watching distance is detected to be continuously lower than a preset safety threshold value, judging that the watching behavior of the child is monitored to violate a preset rule, and controlling the screen brightness to be adjusted according to the violation degree in a linear proportion; And when the watching duration exceeds the set threshold, controlling to trigger screen locking and starting a fourth preset time cooling period.
6. The method for controlling children's anti-addiction based on deep learning and behavior recognition according to claim 1, wherein when the viewing behavior of the children is detected to violate a preset rule, the step of executing the hierarchical intervention reminding and controlling process according to a preset differential gradient adjustment algorithm further comprises: synchronizing all intervention records and behavior data to a parent end App through an MQTT protocol, receiving remote modification and historical data analysis of a threshold value by a parent end operation instruction, and performing personalized parameter setting; And improving and training the YOLOv model identification precision through a federal learning framework, and establishing a light intensity-ranging error compensation model to dynamically correct the ranging value.
7. The method for controlling children's anti-addiction based on deep learning and behavior recognition according to claim 1, wherein after the child identity and the viewing distance are determined, real-time judgment and analysis are performed on the viewing distance, and when the viewing distance is continuously detected to be lower than a preset safety threshold for a predetermined time, the control performs primary early warning, and the step of continuously monitoring the viewing behavior of the current viewing object of the child identity further comprises: Constructing a dual-thread monitoring system to perform real-time analysis on the dynamic behaviors of children, wherein a thread I acquires the data stream of the built-in camera in real time, calculates the watching distance between eyes of a user and a screen, and writes the watching distance into an annular buffer area; Reading viewing distance data from a buffer area by a thread II, controlling to start an accumulation timer when a distance value of a viewing distance detected by continuous preset frames is smaller than a preset safety threshold value, and sending the distance data of the continuous preset frames and corresponding extracted preset three-dimensional feature point coordinate data to a preset gesture stability evaluation module; Analyzing the distance data of the continuous preset frames and the coordinate data of the corresponding extracted preset three-dimensional feature points by the preset gesture stability evaluation module, and determining the relative position change amplitude of the key feature points of the face of the child corresponding to the continuous preset frames; Calculating the angle change of the head of the child in the pitch direction, the yaw direction and the roll direction according to the stability of the head gesture of the child corresponding to the continuous preset frames, and judging whether continuous directional gesture changes exist or not instead of random small-amplitude shaking; When detecting that the average displacement and the maximum displacement of the key feature points are lower than or equal to a preset slight shaking threshold value in the continuous N frames, and the angle change of the head posture change in three directions is smaller than or equal to a preset posture stability threshold value, judging that the currently watched child user is slightly shaking or posture fine-tuning but not actively approaching; when detecting that the average displacement and the maximum displacement of the key feature points are larger than a preset slight shaking threshold value in the continuous N frames, and the angle change of the head gesture change in three directions is larger than a preset gesture stability threshold value, judging that the currently watched child user is actively close to or the gesture change exceeds a preset amplitude, and controlling to enter a sliding window early warning mechanism for judgment; When the sliding window early warning mechanism is started, setting a sliding window early warning mechanism based on a time sequence, carrying out exponential decay weighted average calculation on the viewing distance data by one window unit in a first preset time, and triggering primary early warning when the weighted average value of the continuous preset window units is continuously lower than a safety threshold value.
8. A child anti-addiction control device based on deep learning and behavior recognition, the device comprising: The watching object identity recognition module is used for analyzing the watching object in front of the terminal in real time through a preset improved network model, recognizing whether the current watching object is a child identity or not and eliminating the occasionally-appearing watching object; the viewing distance calculating module is used for calculating the viewing distance between the current viewing object and the screen, which is the identity of the child, by constructing a three-dimensional model of the face of the child and adopting a dynamic calibration flow when the current viewing object is identified as the identity of the child; The early warning judging module is used for judging and analyzing the watching distance in real time based on the determined child identity and the watching distance, and controlling to perform primary early warning when the watching distance is continuously lower than the preset safety threshold value after continuously preset time detection; the hierarchical intervention reminding and processing control module is used for executing hierarchical intervention reminding and control processing according to a preset differential gradient adjustment algorithm when the watching behavior of the child is monitored to violate a preset rule; And the data pushing module is used for synchronizing all the intervention records and the behavior data to the parent end App through the MQTT protocol, receiving remote modification of the threshold value by the operation instruction of the parent end and analysis of the historical data, and carrying out personalized parameter setting.
9. An intelligent terminal comprising a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors, the one or more programs comprising steps for performing the method of any of claims 1-7.
10. A computer readable storage medium, on which a computer program is stored which, when being executed by a processor, enables an electronic device to perform the steps of the method according to any one of claims 1-7.

Description

Deep learning and behavior recognition-based children anti-addiction control method and terminal Technical Field The invention relates to the technical field of computer vision processing, in particular to a child anti-addiction control method and device based on deep learning and behavior recognition, an intelligent terminal and a storage medium. Background Currently, the television watching time and eye health management of children mainly depend on traditional parental control means, such as physical password locking, preset time shutdown function or single distance detection device based on infrared induction. Although the technologies can realize basic protection, the technologies have the defects of insufficient intelligent degree, poor environmental adaptability and the like. The method is characterized in that the infrared sensor cannot accurately distinguish the identity of a user and is easily interfered by ambient light, the timing function lacks a real-time behavior judging mechanism and cannot be dynamically adjusted according to the actual watching posture of a child, and the parent end management function is single and cannot realize remote parameter configuration and data visualization. In addition, the prior art scheme still has technical faults in the aspects of accurate identification of identities, multidimensional behavior analysis, dynamic response mechanisms, intelligent management of parents and the like, and fails to form a complete technical closed loop of detection-analysis-intervention-feedback, so that the problems of high identity misjudgment rate, poor environmental adaptability and incapability of forming closed loop control are caused. Accordingly, there is a need for improvement and development in the art. Disclosure of Invention In order to solve the technical problems, the invention provides a child anti-addiction control method, a device, an intelligent terminal and a storage medium based on deep learning and behavior recognition, which form a complete technical closed loop of detection, analysis, intervention and feedback, and have the advantages of high accuracy of identity recognition for a viewer, strong environmental adaptability, and quick and accurate efficiency of intervention treatment of the behavior of the viewer. The technical scheme of the application is as follows: a children anti-addiction control method based on deep learning and behavior recognition comprises the following steps: Real-time analysis is carried out on the watching object in front of the terminal through a preset improved network model, whether the current watching object is a child identity or not is identified, and the occasionally-appearing watching object is eliminated; When the current watching object is identified as the identity of the child, calculating the watching distance between the current watching object which is the identity of the child and the screen by constructing a three-dimensional model of the face of the child and adopting a dynamic calibration flow; Based on the determined identity and the viewing distance of the child, carrying out real-time judgment and analysis on the viewing distance, and controlling to carry out primary early warning when the continuous preset time detects that the viewing distance is continuously lower than a preset safety threshold value; when the watching behavior of the child is monitored to violate the preset rule, performing grading intervention reminding and control processing according to a preset differential gradient adjustment algorithm. According to the child anti-addiction control method based on deep learning and behavior recognition, wherein the real-time analysis is carried out on the watching object in front of the terminal through a preset improved network model, whether the current watching object is the identity of a child or not is recognized, and the step of eliminating the occasionally-appearing watching object comprises the following steps: The method comprises the steps of collecting a public data set and video images collected by a built-in camera of a terminal in advance, marking the video images, defining a child face area, defining interference objects of adult faces and pets, and constructing a multi-scene child face data set; Carrying out light improvement on the YOLOv model, replacing a standard convolution layer with a phantom network module (GhostNet module), and optimizing a loss function; Analyzing the pictures of the watched objects obtained by the terminal camera in real time through the improved YOLOv model, identifying whether the current watched objects are child identities or not, and eliminating the watched objects which occur occasionally; When the current viewing object is detected as child identity for a continuous predetermined time, it is determined that the current viewing object is identified as child identity. According to the child anti-addiction control method based on deep learning and behavi