Multifunctional camera system for children

Abstract

The invention relates to the technical field of intelligent electronic equipment, and discloses a multifunctional camera system for children, and the system comprises a dynamic resource distribution module which is used for receiving multispectral data and temperature feedback, and generating a processor frequency, an image bandwidth and a memory distribution instruction; the environment sensing module is used for integrating a dual-spectrum sensor, compressing features through a tensor decomposition network and then inputting the features into the resource scheduling evaluation module; the intelligent optimization module interacts with the resource allocation module, dynamically adjusts the reward function weight and receives strategy execution effect data; and the heat management module is used for calculating the maximum allowable frequency based on a heat conduction model and feeding back the temperature gradient data to the resource allocation decision module. Through multispectral data fusion under a reinforcement learning framework, a meta-learning driven multi-objective optimization algorithm and a thermodynamic constrained frequency control technology, collaborative optimization of adaptive resource scheduling, precise temperature control and multi-dimensional safety protection of the child intelligent camera in a complex environment is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent electronic devices, and in particular to a multifunctional camera system for children. Background Art

[0002] With the rapid development of smart educational devices and interactive entertainment products for children, the market is increasingly demanding imaging systems with strong environmental adaptability, high safety protection levels, and intelligent energy management. Traditional camera systems are increasingly limited in their ability to cope with complex lighting conditions, dynamic usage scenarios, and the special safety requirements of children. There is an urgent need to develop new solutions that integrate multimodal perception and intelligent resource scheduling.

[0003] Current mainstream technology solutions often utilize single-channel visible light sensors coupled with fixed-parameter image processing pipelines. Thermal management relies on a stepped frequency reduction strategy based on preset temperature thresholds, and security primarily relies on physical locks and password verification. Resource allocation strategies are typically based on a static task priority table, with optimization focused on improving a single performance metric. Modules exchange data via standardized interfaces.

[0004] However, existing technologies still have some shortcomings: single-spectrum imaging is prone to loss of detail in backlit or low-light environments, crude temperature control with fixed thermal thresholds leads to significant performance fluctuations, and physical protection mechanisms are difficult to deal with sudden drops or unauthorized data access. Furthermore, the discrete subsystems lack coordinated optimization, and resource allocation decisions fail to fully consider the dynamic coupling between environmental conditions and thermodynamic constraints, resulting in low energy efficiency and a fragmented user experience. These shortcomings restrict the reliability and practicality of children's smart devices in complex application scenarios, and breakthroughs through technological innovation are urgently needed. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a multifunctional camera system for children, which solves the problems of rigid resource allocation, conservative temperature control strategy and insufficient safety protection of existing children's smart devices in complex environments.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a multifunctional camera system for children, the system comprising a dynamic resource allocation module, an environmental perception module, an intelligent optimization module, and a thermal management module: The dynamic resource allocation module generates control instructions including processor frequency, image processing bandwidth and memory strategy by receiving multispectral sensing data from the environmental perception module and temperature feedback from the thermal management module; The environment perception module integrates a visible light and near-infrared dual-channel sensor, whose output is compressed by a feature extraction network and then input into the dynamic resource allocation module; The intelligent optimization module forms a two-way data channel with the dynamic resource allocation module. The former provides the dynamic weight parameters of the reward function to the latter, and the latter feeds back resource allocation effect data to the former. The thermal management module receives the frequency control signal output by the dynamic resource allocation module, calculates the maximum allowable operating frequency through an embedded thermodynamic model, and transmits temperature gradient data back to the dynamic resource allocation module.

[0007] The present invention provides a multifunctional camera system for children, which has the following beneficial effects: 1. The present invention achieves accurate perception of scene lighting conditions through the intelligent fusion of visible light and near-infrared spectra and feature tensor compression technology, enabling the dynamic resource allocation strategy to adaptively adjust the image processing bandwidth and computing resource ratio according to the complexity of the environment, maintaining stable imaging quality under complex lighting conditions.

[0008] 2. The present invention constructs a temperature-performance balance curve based on the dynamic frequency limiting mechanism of the embedded heat conduction model, and corrects the processor operating parameters in real time through closed-loop feedback, maximizing the utilization of hardware resources while ensuring the safe operation temperature threshold of the equipment.

[0009] 3. The present invention adopts a hybrid optimization architecture that combines a meta-learning framework with a genetic algorithm, enabling the system to autonomously adjust the weight parameters of the reward function based on historical operating data, and continuously approach the Pareto optimal solution set under multi-dimensional constraints such as energy consumption, image quality, and latency.

[0010] 4. Through modular function integration and hierarchical permission control, a smart interactive ecosystem exclusively for children is built. Communication whitelists and remote authorization mechanisms effectively block unauthorized access. AR games and physical programming interfaces promote cognitive development, while emergency erase and physical locking ensure data security in the event of device loss, achieving an optimal balance between feature richness and security and reliability.

[0011] 5. The present invention realizes the precise matching of computing resource supply and actual demand through the data flow closed-loop design of the environmental perception module, dynamic allocation module and thermal management module, eliminates redundant power consumption while ensuring functional integrity, and extends the single battery life of mobile devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a system structure diagram of the present invention; Figure 2 This is a module architecture diagram of the function extension module of the present invention. DETAILED DESCRIPTION

[0013] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the specification of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0014] Please see the attached Figure 1 and attached Figure 2 An embodiment of the present invention provides a multifunctional camera system for children, which includes a dynamic resource allocation module, an environment perception module, an intelligent optimization module, and a thermal management module: The dynamic resource allocation module generates control instructions including processor frequency, image processing bandwidth, and memory strategy by receiving multispectral sensor data from the environmental perception module and temperature feedback from the thermal management module; In this embodiment, the dynamic resource allocation module implements intelligent scheduling of hardware resources through a reinforcement learning framework. This module establishes a data path with the multispectral environmental perception module and the thermal management module, receiving multi-dimensional sensor data such as ambient light intensity, device motion status, and temperature gradients, and outputs optimized control instructions for processor frequency, image processing bandwidth, and memory management strategies.

[0015] In this embodiment, the state perception subsystem constructs a six-dimensional state vector as a decision basis. The state vector includes the battery energy level , ambient illumination , three-axis acceleration modulus , feature priority vector , chip temperature and storage space status Among them, the battery energy level is normalized using the voltage-capacity conversion curve, the ambient illumination value is logarithmically transformed to adapt to the human eye's perception characteristics, and the function priority vector is dynamically generated according to the user's operation mode, and its dimension is positively correlated with the number of currently activated functions.

[0016] In this embodiment, the action space is designed as a four-tuple control instruction, covering the key parameters of the computing module, image processing and storage resources. Specifically, the CPU operating frequency , graphics processor operating frequency , image signal processor transmission bandwidth, And the memory management policy identifier Preferably, the CPU frequency is adjusted in a discrete step-by-step manner, the GPU frequency and the image signal processor bandwidth are linked and constrained, and the memory management strategy is dynamically switched between the least recently used (LRU) and least frequently used (LFU) algorithms.

[0017] In this embodiment, the policy optimization subsystem uses a temporal difference learning algorithm to implement iterative updates of the Q-value function. The update formula is: ; in, is the learning rate; is the discount factor; is the immediate reward value; is the state vector at the next moment; is the candidate action vector; For the next state The maximum expected Q value under ; Indicates that the status Next action Preferably, the weight coefficients in the reward function are dynamically adjusted through a meta-learning mechanism to adapt to the optimization objective bias in different usage scenarios.

[0018] In this embodiment, the instant reward calculation module integrates multi-source sensor data fusion processing. Battery status reward items and remaining power There is a nonlinear positive correlation, and the temperature rise penalty term is based on the first-order difference of the temperature sensor reading. The image quality reward item receives the feature tensor norm evaluation result output by the environment perception module. Preferably, a sliding time window mechanism is introduced to smooth the instantaneous reward to avoid policy oscillation.

[0019] In this embodiment, an experience replay mechanism ensures the stability of policy learning. The system maintains a fixed-capacity state-action-reward storage buffer, updating training samples in a first-in, first-out manner. Preferably, a prioritized experience replay strategy is employed, dynamically adjusting sample sampling probabilities based on temporal difference error, to prioritize policy learning under unusual conditions.

[0020] In this embodiment, the dual-network architecture of the target network and the policy network improves algorithm convergence. The policy network generates control instructions and collects training data in real time, while the target network regularly synchronizes policy network parameters and calculates the target Q value. Preferably, the target network update period is negatively correlated with the device temperature. In high-temperature conditions, the parameter synchronization interval is extended to reduce computational load.

[0021] In this embodiment, the exploration-exploitation balance mechanism is adaptively Greedy strategy implementation. In the initial stage, a high exploration probability is set to fully traverse the state space, and the utilization ratio is gradually increased as the training progresses.

[0022] In this embodiment, the resource allocation decision establishes a bidirectional constraint relationship with the thermal management module. The frequency control instruction output by the dynamic resource allocation module is limited by the real-time maximum allowable frequency calculated by the thermal management module. When the decision instruction exceeds the safety range, the constraint correction mechanism is triggered. The corrected frequency deviation is fed back to the policy optimization subsystem to adjust the temperature rise penalty weight in the reward function.

[0023] The environmental perception module integrates a visible light and near-infrared dual-channel sensor, whose output is compressed by a feature extraction network and then input into the dynamic resource allocation module; In this embodiment, the environmental perception module achieves intelligent perception of scene information through the synergy of multispectral imaging and deep learning feature extraction. This module establishes a data channel with the dynamic resource allocation module, embedding compressed environmental features into the resource scheduling decision-making process, forming a closed-loop perception-decision-making system.

[0024] In this embodiment, the multispectral acquisition subsystem utilizes a parallel imaging architecture using both visible and near-infrared channels. The visible light channel captures RGB color information, while the near-infrared channel enhances detail in low-light environments. The outputs of the two sensors are timestamped by a hardware synchronization controller to ensure spatiotemporal consistency of the fused data. Preferably, the near-infrared channel is equipped with a tunable filter whose transmission wavelength range adapts to varying lighting conditions.

[0025] In this embodiment, the multispectral data fusion module performs a pixel-level weighted fusion operation, and its calculation expression is: ; in, is the spectral channel identifier; It is the visible light channel identifier; It is the near infrared channel marker; For channel At the pixel The radiation intensity value; is the image plane coordinate; The coordinates of the fused image are The pixel intensity value at ; is the pixel-level fusion weight, and the calculation formula is: ; For channel The local area mean of For channel The local area standard deviation; is a natural exponential function used to convert difference metrics into weight values.

[0026] In this embodiment, the feature extraction subsystem uses tensor decomposition technology to compress the standard ResNet network. The convolution kernel weight matrix is reconstructed through Tucker decomposition as follows: ; in, is the original raw network weight tensor; is the core tensor; For the modal factor matrix; Represents the first Modal product; core tensor Solved by optimization problem: ; in, represents the Frobenius norm; represents the nuclear norm; is the regularization coefficient.

[0027] In this embodiment, the spectral attention mechanism enhances cross-modal feature interaction. A dual-channel association is established through a learnable projection matrix: ; in, is the sigmoid activation function; and They are visible light and near infrared characteristic maps respectively; 、 is the projection parameter matrix. The attention map guides the weighted fusion of features, focusing on preserving cross-modal complementary information.

[0028] In this embodiment, the feature compression network output is deeply coupled with the dynamic resource allocation module. The dimensional information of is encoded into the feature priority vector: ; in, Represents the dimension of the feature priority vector; Decomposing the core tensor for Tucker In the This design enables the resource scheduling strategy to adaptively perceive task complexity and dynamically adjust the computing resource allocation.

[0029] In this embodiment, a multi-scale feature pyramid enhances the robustness of environmental perception. Feature maps are extracted at different depths in the compression network, and multi-resolution feature representations are constructed through upsampling and concatenation. An online calibration module ensures the accuracy of multispectral data. Sensor response calibration is regularly performed using a built-in reference color palette to compensate for temperature drift and device aging.

[0030] The intelligent optimization module forms a two-way data channel with the dynamic resource allocation module. The former provides the dynamic weight parameters of the reward function to the latter, and the latter feeds back resource allocation effect data to the former. In this embodiment, the intelligent optimization module is used to realize the adaptive evolution and optimization of the resource allocation strategy under multi-objective constraints. This module establishes a linkage mechanism with the dynamic resource allocation module, the environmental perception module and the thermal management module. By introducing the meta-learning framework and the multi-objective optimization algorithm, it effectively coordinates the complex balance between performance, energy consumption and temperature control.

[0031] In this embodiment, the intelligent optimization module first implements self-tuning of resource allocation policy parameters by constructing a two-layer meta-learning structure. This structure consists of a base policy network and a meta-optimizer. The base policy network provides current resource scheduling behavior, while the meta-optimizer updates parameters based on cross-task error feedback.

[0032] Meta-learning parameter adjustment: ; in, is the dynamic weight parameter; is the meta-learning rate; is the meta-loss function; Indicates the parameter group The set of partial derivatives of .

[0033] In this embodiment, the meta-loss function LmetaLmeta is used to evaluate the response ability of the policy network to the actual task, and is defined as follows: ; in, Target value; The current policy network in parameters Next Value estimation; is the meta-loss function; is the expectation operator. This loss function drives the parameters to converge towards a better direction by minimizing the estimation error.

[0034] To suppress gradient explosion and improve learning stability, the system introduces a gradient truncation strategy. The calculation process is as follows: ; in, represents the gradient vector of the policy network parameters; is the upper limit of the gradient amplitude; is the truncation function.

[0035] In this embodiment, after meta-learning parameter adjustment is completed, the system enters the multi-objective optimization phase. This phase uses the non-dominated sorting genetic algorithm (NSGA-II) to search the multi-objective solution space. The optimization objectives include system energy consumption, image quality, chip temperature rise, and processing delay.

[0036] The multi-objective optimization function is defined as follows: ; in, is the total energy consumption of the system; is the image quality evaluation value; The real-time temperature of the chip surface; is the maximum allowable operating temperature; Image processing pipeline latency; The maximum allowed delay threshold.

[0037] In this embodiment, the nonlinear constraints in the system optimization problem are processed using a penalty function, as follows: ; in, is the decision variable vector of the optimization problem; is the total number of constraints; Indicates the inequality constraint functions; Indicates that penalties are only imposed for violations of the constraints.

[0038] In terms of solver implementation, NSGA-II configures steps such as population initialization, non-dominated sorting, crowding calculation, and elite strategy retention. It generates new individuals through crossover and mutation operations, and approaches the Pareto optimal frontier through multiple generations of iteration. Optimally, the population size and number of evolutionary generations can be dynamically adjusted based on the system's real-time computing power.

[0039] In this embodiment, in order to enhance the practicality of the optimization results, the system weights and scores the individuals in the Pareto solution according to the target priority in the current usage scenario, and selects the resource allocation strategy with the highest score as the final output. The scoring mechanism is provided by the dynamic resource allocation module with the priority vector , embedded in the optimizer objective function evaluation to guide the search direction.

[0040] In addition, in this embodiment, the intelligent optimization module establishes a periodic interaction mechanism with other system modules. The meta-learning process is triggered once every fixed time period by default, and the optimization process dynamically adjusts the trigger frequency based on environmental changes.

[0041] In this embodiment, the intelligent optimization module also features a policy rollback mechanism. If a new policy results in a significant drop in reward value during actual operation, the system will automatically roll back to the previous Pareto optimal solution and record the failed policy as a negative example in the meta-learning dataset to prevent it from being repeatedly sampled in the future.

[0042] The thermal management module receives the frequency control signal output by the dynamic resource allocation module, calculates the maximum allowable operating frequency through an embedded thermodynamic model, and transmits the temperature gradient data back to the dynamic resource allocation module; In this embodiment, the thermal management module achieves safe device temperature control through the synergy of thermodynamic modeling and dynamic frequency control. This module establishes a bidirectional data channel with the dynamic resource allocation module, receiving processor operating parameters and providing feedback on temperature constraints, forming a closed-loop temperature control system.

[0043] In this embodiment, the thermodynamic modeling subsystem constructs a chip temperature field prediction model based on the three-dimensional heat conduction equation. Its control equation is expressed as: ; in, is the thermal diffusivity; is the specific heat capacity; is the material density; is the heat dissipation power, calculated by ; is the operating voltage; The model comprehensively considers conduction, convection, and heat source terms to accurately describe the temperature distribution inside the chip.

[0044] In this embodiment, the finite element discretization method converts the continuous heat equation into grid node temperature calculation. In the grid module, the temperature of each node is updated as follows: ; in, is the time step; is the spatial step length; represents the set of adjacent nodes; Indicates the The grid node in Temperature value of each time step; is the thermal diffusivity of the material; For the Heat dissipation power per node; is the specific heat capacity of the material; Preferably, an adaptive time step adjustment strategy is adopted to automatically reduce the calculation step size when the temperature changes drastically.

[0045] In this embodiment, the boundary condition processing module supports modeling of various heat dissipation scenarios. The upper surface adopts a mixed convection-radiation boundary: ; in, is the convective heat transfer coefficient; is the surface emissivity; is the Stefan-Boltzmann constant; is the ambient temperature; Represents the gradient of the temperature field in the direction of the boundary normal; is the boundary surface temperature.

[0046] In this embodiment, the dynamic frequency control subsystem adjusts the processor operating upper limit in real time based on the temperature prediction result. The maximum allowable frequency calculation follows: ; in, is the temperature difference between the chip and the environment; is the characteristic parameter of the shell material; is the reference frequency; The maximum frequency allowed by the thermal management module is calculated. Preferably, the reference frequency mapping table is stored in a non-volatile memory to support OTA remote updates.

[0047] In this embodiment, the temperature-frequency mapping table establishes a multi-level safety threshold. When the real-time temperature enters different ranges, the hierarchical frequency reduction strategy is triggered: Safe zone: maintain base frequency operation; Warning zone: frequency reduction is carried out smoothly according to the exponential curve; Dangerous Zone: Forced Switching to Safe Mode This design ensures equipment safety while maximizing performance output.

[0048] In this embodiment, the frequency smooth transition mechanism avoids sudden changes in the operating state. A first-order inertia link is used to achieve frequency gradual change: ; in, is the adjusted processor frequency; is the current operating frequency; is the target frequency; is the proportional adjustment coefficient; Dynamically adjust according to the temperature change rate. When a sharp rise in temperature is detected, the value to speed up the frequency reduction response.

[0049] In this embodiment, the modified feedback mechanism ensures that resource allocation decisions comply with thermal constraints. Exceed When forced correction is performed: ; in, is the corrected CPU operating frequency; The CPU frequency instruction value originally output by the dynamic resource allocation module; The maximum allowable frequency calculated by the thermal management module; 0.9 is the safety margin factor.

[0050] Correction amount Feedback is fed back to the intelligent optimization module to adjust the temperature rise penalty weight in the reward function. This closed-loop design achieves deep coupling between temperature control strategy and resource scheduling.

[0051] In this embodiment, the thermal history module stores temperature time series data, providing training samples for the optimization algorithm. A sliding window mechanism is used to extract feature vectors, including metrics such as average temperature rise rate, peak temperature duration, and frequency adjustment frequency. These features contribute to the construction of the meta-learning state space, enhancing the strategy's adaptability to thermal scenarios.

[0052] In this embodiment, enhanced cooling mode is activated when persistent high temperatures are detected. This collaborative cooling strategy is implemented through a variety of measures, including increasing fan speed, adjusting task scheduling strategies, and disabling non-core functions. Preferably, a cooling strategy selector dynamically selects the optimal combination based on the temperature exceeding the specified limit and the battery status.

[0053] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A multifunctional camera system for children, characterized in that: The system includes a dynamic resource allocation module, an environmental perception module, an intelligent optimization module, and a thermal management module: The dynamic resource allocation module generates control instructions including processor frequency, image processing bandwidth and memory strategy by receiving multispectral sensing data from the environmental perception module and temperature feedback from the thermal management module; The environment perception module integrates a visible light and near-infrared dual-channel sensor, whose output is compressed by a feature extraction network and then input into the dynamic resource allocation module; The intelligent optimization module forms a two-way data channel with the dynamic resource allocation module. The former provides the dynamic weight parameters of the reward function to the latter, and the latter feeds back resource allocation effect data to the former. The thermal management module receives the frequency control signal output by the dynamic resource allocation module, calculates the maximum allowable operating frequency through an embedded thermodynamic model, and transmits temperature gradient data back to the dynamic resource allocation module.

2. The multifunctional camera system for children according to claim 1, characterized in that: The dynamic resource allocation module performs the following process: Constructing a six-dimensional state vector ,in, : Normalized battery capacity; is the ambient illumination; is the triaxial acceleration modulus; is the function priority vector; Read the temperature sensor; The remaining ratio of storage space; Generate control instructions ,in, is the CPU frequency; is the GPU frequency; ISP bandwidth; Memory policy identifier; Update the strategy through the Q-learning algorithm: ； in, is the learning rate; is the discount factor; is the immediate reward value; is the state vector at the next moment; is the candidate action vector; For the next state The maximum expected Q value under ; Indicates that the status Next action The expected cumulative reward value of .

3. The multifunctional camera system for children according to claim 1, characterized in that: The environment perception module performs the following process: Multispectral data fusion: ； in, is the spectral channel identifier; It is the visible light channel identifier; It is the near infrared channel marker; For channel At the pixel The radiation intensity value; is the image plane coordinate; The coordinates of the fused image are The pixel intensity value at ; is the pixel-level fusion weight, and the calculation formula is: ； For channel The local area mean of For channel The local area standard deviation; is a natural exponential function used to convert difference metrics into weight values; Feature extraction network compression: ResNet18 is converted to: ； in, is the original raw network weight tensor; is the core tensor; For the modal factor matrix; Represents the first Modal product; core tensor Solved by optimization problem: ； in, represents the Frobenius norm; represents the nuclear norm; is the regularization coefficient.

4. The multifunctional camera system for children according to claim 3, characterized in that: The feature extraction network output is associated with the state vector of the dynamic resource allocation module in the following way: Compress the feature tensor The dimensional information is encoded into the function priority vector ,satisfy: ; in, Represents the dimension of the feature priority vector; Decomposing the core tensor for Tucker In the Dimensions.

5. The multifunctional camera system for children according to claim 1, characterized in that: The intelligent optimization module performs the following process: Meta-learning parameter adjustment: ； in, is the dynamic weight parameter; is the meta-learning rate; is the meta-loss function; Indicates the parameter group The set of partial derivatives of ; Meta-loss function: ； in, Target value; The current policy network in parameters Next Value estimation; is the meta-loss function; is the expectation operator; Multi-objective optimization solution: ； in, is the total energy consumption of the system; is the image quality evaluation value; The real-time temperature of the chip surface; is the maximum allowable operating temperature; Image processing pipeline latency; The maximum allowed delay threshold.

6. The multifunctional camera system for children according to claim 1, characterized in that: The thermal management module performs the following process: Thermodynamic modeling: ； in, is the thermal diffusivity; is the specific heat capacity; is the material density; is the heat dissipation power, calculated by ; is the operating voltage; is the thermal resistance; Dynamic Frequency Limiting: ； in, is the temperature difference between the chip and the environment; is the characteristic parameter of the shell material; is the reference frequency; The maximum allowed frequency calculated for the thermal management module.

7. The multifunctional camera system for children according to claim 6, characterized in that: The control instructions of the dynamic frequency limit and dynamic resource allocation modules interact in the following ways: when When forced to correct the CPU frequency: ； in, is the corrected CPU operating frequency; The CPU frequency instruction value originally output by the dynamic resource allocation module; The maximum allowable frequency calculated by the thermal management module; 0.9 is the safety margin factor.

8. The multifunctional camera system for children according to claim 1, characterized in that: The system also includes a function expansion module, including: The communication submodule supports preset contact whitelists and parent-authorized calls; The educational tool submodule integrates image recognition popular science and e-book readers; Entertainment sub-module, with built-in AR puzzle game and physical building block programming interface; The security control submodule provides emergency physical lock and automatic data erasure functions.

Images