CN-121980580-A - Astrocyte memristor network driven time sequence analysis and dynamic safety method, system, equipment and medium

CN121980580ACN 121980580 ACN121980580 ACN 121980580ACN-121980580-A

Abstract

The invention discloses a time sequence analysis and dynamic safety method, a system, equipment and a medium driven by an astrocyte memristor network, which belong to the technical field of computer vision and information safety and comprise the steps of collecting video streams and extracting human skeleton sequences; the method comprises the steps of carrying out space-time alignment evaluation on a skeleton sequence and a standard template, identifying abnormal actions, constructing a quadruple synapse network comprising Chay neurons, HR neurons, astrocytes and memristors, dynamically adjusting electric coupling parameters through gesture recognition to generate chaotic signals, extracting multidimensional features from the chaotic signals, generating key streams through hash enhancement and key expansion, encrypting video data by using the key streams, and generating scoring suggestions. According to the invention, lightweight real-time encryption is realized through the biological inspired chaotic neural network, the problem of high computational complexity of the traditional encryption algorithm is solved, the balance of instantaneity and safety is realized, the gesture dynamic control is supported, and the method is suitable for the scenes of rehabilitation training, posture correction and the like.

Inventors

GAN YONGJUN
LI FUDONG
LV YILIN
YANG YAHUI
ZHANG HAIMING
ZHU ZHIHAO
Luo Luxun

Assignees

北京信息科技大学

Dates

Publication Date: 20260505
Application Date: 20251225

Claims (10)

1. A time sequence analysis and dynamic safety method for astrocyte memristor network driving is characterized by comprising the following steps of, Collecting a user video stream through a camera, and extracting a human body key point skeleton sequence based on a gesture estimation algorithm; Performing space-time alignment and similarity evaluation on the skeleton sequence and a standard template, and identifying an abnormal action period; Establishing a quadruple synaptic network based on Chay neurons and HR neurons, dynamically adjusting electric coupling parameters through gesture recognition, enabling the quadruple synaptic network to enter a synchronous state and generating chaotic signals; extracting multidimensional chaotic characteristics from the chaotic signal, and generating a key stream through the reinforcement of the cryptographic hash and the expansion of a key; encrypting the video data using the key stream; a scoring suggestion is generated.
2. The method for astrocyte memristor network driven time sequence analysis and dynamic security according to claim 1, wherein the extracting human body key point skeleton sequence based on the posture estimation algorithm comprises the following steps: Extracting normalized coordinate sequences of 33 human body key points; Coordinate translation of the pelvic center and scaling based on torso length; based on the normalized joint coordinates, defining an inter-joint Euclidean distance and an inter-bone included angle; And integrating the Euclidean distance between joints and the included angle between bones into a space geometrical feature vector according to frames.
3. The method for astrocyte memristor network driven timing analysis and dynamic security as set forth in claim 2, wherein the performing the space-time alignment and similarity assessment on the skeleton sequence and the standard template, and the identifying the abnormal action period comprises: Constructing a space-time coupling feature vector, wherein the space-time coupling feature vector comprises the space geometric feature vector and a time feature; Defining weighted space-time Euclidean distance as an inter-frame similarity measure based on the space-time coupling feature vector; Calculating an accumulated cost matrix based on the weighted space-time Euclidean distance under the partial slope constraint and the joint topological constraint by adopting a dynamic time warping algorithm, wherein the partial slope constraint limits the bending factor of an alignment path, and the joint topological constraint divides the joint points into functional groups and preferentially aligns joints of the same functional group; And identifying abnormal action time periods according to the accumulated cost matrix.
4. The method for astrocyte memristive network-driven timing analysis and dynamic security of claim 3, wherein the constructing the Chay neuron-and HR neuron-based quadruple synaptic network comprises: Establishing a quadruple-coupled differential equation set model comprising Chay neurons, HR neurons, astrocytes and memristors electrically coupled; acquiring a gesture type through gesture recognition, and adjusting electrochemical coupling strength parameters in the quadruple coupling differential equation set model according to the gesture type; And solving the quadruple coupling differential equation set model to obtain membrane potential time sequences of the Chay neurons and the HR neurons, and outputting chaotic signals when the membrane potential time sequences enter a synchronous state.
5. The method for astrocyte memristive network driven timing analysis and dynamic security as set forth in claim 4, wherein generating the keystream comprises: Extracting multidimensional chaotic characteristics from the chaotic signal, wherein the multidimensional chaotic characteristics comprise membrane potential symbol characteristics, membrane potential difference symbol characteristics, phase relation characteristics and neuron-crossing membrane potential comparison characteristics; Interleaving the multi-dimensional chaotic features into a combined bit stream; selecting the coupling parameters of the quadruple coupling differential equation set model as salt values, and carrying out hash processing on the combined bit stream and the salt values by using a hash algorithm; And performing key expansion on the hash processing result based on a key derivation function to generate the key stream.
6. The method for astrocyte memristive network driven timing analysis and dynamic security as set forth in claim 5, wherein the encrypting the video data using the keystream comprises: dividing each frame of image data of a video stream into a plurality of data blocks; Calculating an encryption offset according to the frame sequence number and the time stamp; and respectively carrying out stream cipher encryption on the plurality of data blocks by using different sections corresponding to the encryption offset in the key stream.
7. The method of astrocyte memristive network driven timing analysis and dynamic security of claim 6, wherein generating the scoring advice comprises: calculating the overall action similarity score based on the accumulated cost matrix, and identifying the position of the abnormal action period on a time axis; extracting joint angle deviation and movement speed difference of the abnormal action period; based on the joint angular deviation and the motion velocity difference, a semantic correction proposal is generated.
8. An astrocyte memristor network driven timing analysis and dynamic security system, applying the astrocyte memristor network driven timing analysis and dynamic security method as set forth in any one of claims 1 to 7, comprising: the gesture extraction module is used for acquiring a user video stream through a camera and extracting a human body key point skeleton sequence based on a gesture estimation algorithm; the evaluation module is used for carrying out space-time alignment and similarity evaluation on the skeleton sequence and the standard template and identifying abnormal action time periods; The neural network module is used for constructing a quadruple synaptic network based on Chay neurons and HR neurons, dynamically adjusting electric coupling parameters through gesture recognition, enabling the quadruple synaptic network to enter a synchronous state and generating chaotic signals; The key generation module is used for extracting multidimensional chaotic characteristics from the chaotic signal and generating a key stream through the reinforcement of the cryptographic hash and the expansion of the key; An encryption module for encrypting the video data using the key stream; And the feedback module is used for generating scoring suggestions and correcting suggestions.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of a time-series analysis and dynamic security method of astrocyte memristive network actuation of any one of claims 1 to 7.
10. A computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor performs the steps of an astrocyte memristive network driven timing analysis and dynamic security method of any one of claims 1 to 7.

Description

Astrocyte memristor network driven time sequence analysis and dynamic safety method, system, equipment and medium Technical Field The invention relates to the technical field of computer vision and information security, in particular to a time sequence analysis and dynamic security method, a system, equipment and a medium for driving an astrocyte memristor network. Background Currently, mainstream health care and fitness equipment (such as Holomotion) in the market mostly adopts a visual sensor and a deep learning technology to realize motion capture and assessment, but has defects in data security, privacy protection, interaction experience and system efficiency. In terms of data security, after the existing system finishes gesture estimation and action estimation, video streams, skeleton sequences and estimation results of users are often stored and transmitted in an unencrypted form, and the data contain biological characteristic information of the users and have data leakage risks. In order to solve the problem, the traditional encryption algorithms such as AES, RSA and the like are introduced into part of the system, but the algorithm has higher calculation complexity, can generate resource competition with real-time visual processing when running on an embedded platform, and the system needs to balance between the safety intensity and the response speed, so that the requirements of clinical rehabilitation training on high frame rate and low delay are difficult to meet. In the aspect of identity authentication, the existing system mostly adopts static passwords or fixed gestures to carry out user authentication, and the mode lacks dynamic adaptability, and once authentication credentials are revealed, the threat of unauthorized access can be continuously faced. The rehabilitation evaluation data generated by the system also generally lacks an integrity protection and trusted storage mechanism, and when a user forgets to authenticate credentials, a safe recovery path is lacking, so that the long-term rehabilitation process is easily interrupted. From the technical architecture, the existing system has the following two problems: first, there is a lack of a biologically inspired lightweight security mechanism. The encryption module and the action evaluation module in the existing system are independent from each other, and the biological dynamics characteristics (such as nonlinearity and time sequence characteristics of neuron activity) of the rehabilitation action are not utilized to generate a secret key, so that a security mechanism is provided with independent calculation burden. Second, key generation is disjointed from the user action state. The key of the existing system is usually pre-generated or statically configured, is irrelevant to the rehabilitation action quality, rhythm and physical state of the user, and cannot utilize the random source naturally generated in the rehabilitation process. In summary, in the prior art, the visual perception module and the security encryption module are usually designed separately, and lack of deep fusion of an algorithm layer and a data stream layer, so that the system is difficult to consider real-time performance, security and user experience in an embedded environment. Disclosure of Invention In view of the above-mentioned problems, the present invention provides a method, a system, a device and a medium for timing analysis and dynamic security of astrocyte memristor network driving. Therefore, the invention solves the technical problems that firstly, the problem of lack of real-time lightweight encryption protection of sensitive biological data in a vision rehabilitation system is required to be solved. The existing system usually processes visual perception and data security as two independent modules, so that an effective encryption mechanism is difficult to embed in the real-time video stream analysis process, and sensitive biological data such as a skeleton sequence, a motion track and an evaluation result of a user face leakage risks. Secondly, the fundamental contradiction between the traditional encryption algorithm and the real-time requirement of the embedded system must be broken through. The high computational complexity of the traditional encryption algorithm and the limited computational power of the embedded equipment form direct conflict, so that the system is always forced to make a compromise between the safety intensity and the response speed, and the clinical requirement of rehabilitation training on real-time feedback cannot be met. Again, there is a need to change the authentication mechanism that is stiff in existing systems. The current system relies on static passwords or fixed gestures for identity authentication, and the interaction mode of the system and the rehabilitation process is not only lack of natural experience, but also has potential safety hazards that the system is completely exposed once leakage occurs. Finally, a