An interactive physical experiment video analysis system and method based on generative artificial intelligence
By explicitly declaring parameters and dependency graphs using generative artificial intelligence, dynamic parameter adjustment and real-time response of the physical experiment video analysis system are realized, solving the flexibility and efficiency problems of existing systems and supporting unified processing of live footage and simulation analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BAOJI UNIV OF ARTS & SCI
- Filing Date
- 2026-01-27
- Publication Date
- 2026-06-02
AI Technical Summary
Existing physics experiment video analysis systems suffer from difficulties in dynamically adjusting analysis parameters, low computational efficiency, lack of flexibility, poor real-time interactive effects, difficulty in unifying the processing of live footage and simulation analysis, and high costs for expansion and customization.
Generative artificial intelligence is used to generate video analysis program code with explicit parameter declarations, establish a dependency graph between parameters and calculation steps, realize selective recalculation and real-time visualization, and support natural language interaction and parameter adjustment.
It improves the controllability and interpretability of parameters in the analysis process, enhances real-time response performance, reduces expansion and customization costs, and supports unified processing of real-world and simulation analyses.
Smart Images
Figure CN122132013A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer-aided experiment and intelligent interactive analysis technology, specifically to a human-computer interactive physical experiment video analysis system and method that integrates generative artificial intelligence, parametric computation and real-time visualization. Background Technology
[0002] With the development of computer vision and digital video processing technologies, video-based physical experiment analysis systems are widely used in scenarios such as moving target recognition, trajectory extraction, and physical quantity calculation to obtain kinematic and dynamic information of experimental objects from video data. These systems typically perform preprocessing, target detection, trajectory analysis, and model calculation on the video to output analytical results such as displacement, velocity, acceleration, momentum, or energy.
[0003] However, existing physics experiment video analysis systems generally employ fixed analysis workflows and static parameter configurations. Core analysis parameters (such as image processing thresholds, target detection conditions, and physical model parameters) are typically determined during program initialization and are difficult to adjust dynamically during operation. When users need to change parameters, they often have to restart the analysis workflow or execute all calculation steps, resulting in a lack of flexibility, low computational efficiency, and difficulty in supporting interactive parameter adjustment requirements.
[0004] Furthermore, most systems do not explicitly model the dependencies between analysis parameters and specific calculation steps. When parameters change, the system typically recalculates the entire process, failing to reuse unaffected intermediate calculation results. This can easily cause response delays when processing high-resolution videos or complex models, limiting the real-time interactive experience.
[0005] On the other hand, when customizing functions for different experimental scenarios, existing systems often rely on manual script writing or modification. Parameter definition, interface interaction and calculation logic are tightly coupled, lacking a unified parameter declaration and automatic parsing mechanism, resulting in high costs for expansion and customization.
[0006] Meanwhile, data analysis based on real-time video and simulation analysis based on physical models usually use different processing architectures. They lack a unified mechanism in terms of parameter system, calculation process and result presentation, making it difficult to achieve comparative verification under the same framework, which increases the complexity of the system. Summary of the Invention
[0007] To address the problems in existing technologies, such as difficulty in dynamically adjusting analysis parameters, low computational efficiency when parameters change, poor interpretability of the analysis process, and difficulty in coordinating real-world and simulation analysis, this invention aims to provide an interactive video analysis system and method that can decouple the analysis process through explicit parameter declaration and perform selective recalculation based on parameter dependencies, thereby improving the system's real-time response performance, analysis transparency, and scalability.
[0008] To address the aforementioned technical problems, this invention provides an interactive physics experiment video analysis system based on generative artificial intelligence, comprising: The code generation module is used to receive a functional description in natural language and generate executable video analysis program code, wherein at least one adjustable analysis parameter is explicitly declared in the program code according to predefined parameter declaration rules; A parametric analysis engine, used to load and run the video analysis program code, includes: The parameter dependency resolution unit is used to establish a dependency graph between the adjustable analysis parameters and the specific analysis and calculation steps in the video analysis process when the program code is running. The computation scheduling unit is used to, upon receiving an update instruction for any of the adjustable analysis parameters, trigger recalculation only for analysis calculation steps that have a direct or indirect dependency relationship with the parameter, based on the dependency graph. The interactive interface module is used to dynamically parse the adjustable analysis parameters and their metadata declared in the program code at runtime, and automatically generate and present the corresponding graphical adjustment controls according to the data type of the parameters and the metadata constraints. The real-time visualization module is used to receive the updated analysis results output by the parameterized analysis engine after the value of the adjustable analysis parameter changes, and to perform real-time visualization rendering and output. The data recording and analysis module is used to automatically record different parameter settings and their corresponding analysis results, and supports the generation of correlation analysis charts between parameters and results based on the recorded data.
[0009] Preferably, the code generation module is specifically used to: based on the natural language function description, call a large language model and combine it with a preset physical experiment analysis code template to generate structured video analysis program code; the code template defines a general analysis process framework and parameterizable interface for different types of physical experiments.
[0010] Preferably, the parametric analysis engine is configured to support unified processing of at least two data source types, including: a) Externally imported live-action physics experiment video data; b) Simulation video data is calculated and generated in real time by the system's built-in physical simulation engine based on the user-defined physical model parameters.
[0011] An interactive physics experiment video analysis system based on generative artificial intelligence, characterized in that the adjustable analysis parameters include at least one of the following categories: a) Image processing parameters, including color threshold, filter kernel size, and contour detection sensitivity for target recognition; b) Kinematic analysis parameters, including the starting frame for trajectory extraction, sampling interval, and smoothing window size; c) Dynamic model parameters, including the object's mass, elastic restitution coefficient, damping coefficient, and gravitational acceleration.
[0012] Preferably, when the computation scheduling unit triggers a recalculation, it automatically identifies and reuses all intermediate computation results in the current analysis process that are not affected by the updated parameters.
[0013] Preferably, the real-time visualization module is configured to output at least two visualization formats simultaneously, including: the original video footage with motion trajectory markers and physical quantity annotations superimposed, a physical quantity curve graph that changes over time, and a dynamic display panel of key calculation result values.
[0014] Preferably, the graphical adjustment controls generated by the interactive interface module include one or more combinations of sliders, numerical input boxes, drop-down selection boxes, and switch buttons.
[0015] This invention also provides a method for analyzing interactive physical experiment videos based on generative artificial intelligence, comprising: Responding to the functional description in natural language form, the code generation module generates video analysis program code containing explicit declarations of adjustable analysis parameters; The video analysis program code is loaded and run by the parametric analysis engine to perform analysis and calculation on the input video data. At the same time, the parameter dependency parsing unit establishes the dependency relationship between parameters and calculation steps, and the interactive interface module automatically generates parameter adjustment controls. The interactive interface module receives dynamic adjustment operations from the user on at least one of the adjustable analysis parameters. In response to the adjustment operation, the computation scheduling unit of the parameterized analysis engine selectively triggers the recalculation of related analysis and computation steps based on the dependency relationship, and drives the real-time visualization module to update the output. The data recording and analysis module stores the current parameter combination and the corresponding analysis results.
[0016] Preferably, the "selective triggering of recalculation of associated analysis and calculation steps" specifically includes: locating the affected node in the dependency graph according to the parameter update type, executing only the downstream calculation chain starting from that node, and skipping the calculation of upstream and other unrelated branches.
[0017] Preferably, the method further includes: in the data recording and analysis module, comparing the analysis results under multiple parameter settings recorded according to user selection, and automatically generating a curve or relationship chart reflecting the impact of parameter changes on the results.
[0018] Compared with related technologies, the fully automatic water level and temperature recorder provided by this invention has the following advantages: By explicitly declaring and dynamically exposing analysis parameters, the closed, static processing flow is transformed into an interpretable and adjustable interactive analysis flow, significantly improving the parameter controllability and interpretability of the analysis process.
[0019] By establishing the dependency relationship between parameters and calculation steps, only the relevant steps are recalculated and intermediate results are reused when parameters change, which effectively avoids repeated calculations throughout the process and greatly improves the real-time response performance in interactive parameter adjustment scenarios.
[0020] By adopting a unified parameter declaration, automatic parsing, and control generation mechanism, the analysis logic, parameter configuration, and interface interaction are decoupled, reducing the cost of customizing and expanding video analysis tools for different physical experiment scenarios.
[0021] It supports processing real-world experimental videos and physical model-based simulation videos under the same parametric analysis workflow, which facilitates parameter consistency control, result comparison and model verification, and improves the overall uniformity and applicability of the system. Attached Figure Description
[0022] Figure 1 This is a schematic diagram of the system module structure and data flow of the present invention.
[0023] Figure 2 This is a schematic diagram of the interactive interface layout of the present invention (the left side is the parameter control area, and the right side is the visualization area).
[0024] Figure 3 This is a schematic diagram of the accuracy of the system's automatic experiment (taking gravity acceleration fitting as an example).
[0025] Figure 4 This is a schematic diagram of the parameter-result relationship curve automatically generated by the system (taking the recovery coefficient-energy loss rate as an example).
[0026] Figure 5 A schematic diagram of the visualization results of the free fall experiment (including video tracking, displacement-time curve, and real-time display of g value).
[0027] Figure 6 A schematic diagram illustrating the visualization results of a student's free fall experiment.
[0028] Figure 7This is a schematic diagram of the collision experiment analysis results (including collision animation, momentum change curve, and parameter adjustment controls).
[0029] Figure 8 This is a schematic diagram of the results of the chaotic motion analysis of the double pendulum (including motion animation, phase space trajectory, and Lyapunov exponential curve). Detailed Implementation
[0030] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0031] This invention provides an interactive physics experiment video analysis system based on generative artificial intelligence. Its core modules include a code generation module, a parametric analysis engine, an interactive interface module, a real-time visualization module, and a data recording and analysis module. Furthermore, this invention also provides an interactive physics experiment video analysis method based on this system.
[0032] First, the system architecture and core mechanisms will be explained. For example... Figure 1 As shown, the system constructs a closed-loop process of "natural language customization - parameter interaction - real-time feedback - data accumulation".
[0033] Detailed Explanation of the System's Core Mechanism Template-guided generation mechanism of code generation module At the heart of this module lies a structured, extensible library of physical experiment analysis templates. Each template is a semi-finished code framework containing a three-layer structure: Meta-information layer: Defines the template's unique identifier, experiment category, list of keywords for matching, and natural language description.
[0034] The process skeleton layer defines the immutable sequence of standard analysis steps for this type of experiment. These steps are logically related in the form of a directed acyclic graph. For example, the fixed skeleton of a general kinematic analysis template is: "Video frame decoding and preprocessing" -> "Coordinate system calibration and perspective correction" -> "Target detection and tracking" -> "Trajectory data sequence extraction" -> "Kinematic physical quantity calculation" -> "Model fitting and parameter output".
[0035] Parameterizable interface points: At each step node of the process skeleton, predefined interface points marked with specific annotations are provided. Each interface point is associated with metadata, including parameter name, type, default value, value range, and description text.
[0036] When a functional description in natural language is received, the module's workflow is as follows: (1) Intent Understanding and Template Retrieval: A lightweight text embedding model is used to convert user commands into vectors, and semantic similarity is calculated between the vectors and the metadata description vectors of all templates in the template library. The top N candidate templates with the highest similarity are returned. Internal testing shows that the top-3 hit rate of this method exceeds 99%.
[0037] (2) Structured code generation: The process skeleton and interface point information of the best matching template are used as immutable constraints, and together with user instructions, a structured prompt is constructed and submitted to the large language model. The prompt instructs the large language model to fill in the specific algorithm implementation code within the skeleton framework, and to bind the adjustable variables that the user cares about to the predefined interface points or create new explicit parameter declarations. Using this "template-guided" method, the success rate of generating directly runnable code has increased from about 40% to over 90%.
[0038] (3) Code standardization and security review: The generated code needs to be statically analyzed to ensure that the parameter declaration format is uniform (e.g., placed in the beginning of the file and using the @param decorator), and to filter out code patterns that may have security risks.
[0039] Dependency resolution and incremental calculation mechanism of the parametric analysis engine This engine is the core of the system's real-time interactive performance, and its workflow is as follows: (1) Dependency Graph Construction: When the generated video analysis program code is loaded and run, the engine injects a lightweight runtime tracer. When the output of a function or computation unit (denoted as A) in the code is called as input by another unit (denoted as B), the tracer automatically establishes a directed dependency edge from A to B in memory. All explicitly declared adjustable parameters serve as input source nodes for the entire computation graph. Finally, the system dynamically constructs a complete computation dependency graph in memory.
[0040] (2) Selective Recalculation Scheduling: When a user adjusts a parameter P through the interface, the computation scheduling unit is triggered. It first marks the node corresponding to parameter P as "dirty" in the dependency graph. Then, it executes a graph traversal algorithm, recursively marking all downstream computation nodes that directly or indirectly depend on the "dirty" node as "dirty." Finally, the scheduling unit only initiates the execution of computation tasks corresponding to all "dirty" nodes in the graph, skipping all nodes that are still in a "clean" state and directly reusing their existing computation results. For example, in a free-fall experiment, adjusting the "starting frame offset" parameter only makes the "trajectory extraction" and its downstream "velocity calculation" and "g-value fitting" nodes dirty, while upstream nodes such as "video decoding" and "color calibration" are reused. Experimental results show that this incremental computation strategy can reduce the recalculation time after parameter adjustment to less than 30% of the total computation time, achieving millisecond-level response.
[0041] Automatic control generation and binding mechanism for interactive interface modules This module decouples the analysis logic from the user interaction.
[0042] (1) Parameter declaration parsing: During system initialization, the interface module parses the parameter declaration block in the unified format at the beginning of the video analysis program code. This block lists the name, data type, value range, step size, default value, and display name of all adjustable parameters in a machine-readable form.
[0043] (2) Dynamic generation of controls: Based on the parsed parameter metadata, the module calls the corresponding UI component factory method. For example, for floating-point parameters, a slider control is generated and the minimum, maximum and step values are automatically set; for enumeration parameters, a drop-down selection box is generated and the options are filled. All controls are automatically laid out in the parameter control area of the interactive interface.
[0044] (3) Dynamic event binding: Set a value change listener for each control. When the user interacts with the control, the listener callback function does not directly contain complex business logic, but instead sends a lightweight parameter update message to the parameterized analysis engine, containing the parameter ID and the new value, thereby triggering the aforementioned incremental calculation process.
[0045] Collaboration between the real-time visualization module and the data logging module The real-time visualization module subscribes to the output stream of the parametric analysis engine. Once the engine completes a selective recalculation, the module immediately acquires the new result data and drives the graphical interface to update the video overlay layer, redraw the data curves, and refresh the numerical display panel.
[0046] The data recording and analysis module monitors all parameter change events and corresponding final analysis results in the background. It automatically stores each set of "parameter combination-timestamp-result data" into a structured experimental database. Users can request the system to generate two-dimensional relationship curves or comparison charts based on historical data, such as "parameter A-result B," at any time via the interface. Figure 4 As shown.
[0047] Through the synergistic operation of the aforementioned core mechanisms, this invention achieves a transformation from static analysis to dynamic interaction. The specific applications and technical effects of this invention are further illustrated below through three embodiments.
[0048] Example 1: Investigation of Free Fall Acceleration (Based on Real-World Video) This embodiment uses the "free fall motion" experiment in high school physics as a scenario to demonstrate how the system processes real-time video and enables interactive exploration of parameters.
[0049] Students use their mobile phones to shoot a video containing a vertical ruler and a freely falling red ball and upload it to the system. Teachers or students input natural language commands into the system: "Generate a free-fall real-life video analysis tool that can correct video tilt, automatically track the trajectory of the red ball, calculate gravitational acceleration g through quadratic fitting, allow adjustment of the starting frame offset, rotation correction angle, and ruler width, and display the trajectory diagram and g value in real time." The code generation module responds to instructions, matches and invokes the "free fall analysis template," guiding the large language model to generate specific analysis code. The code includes algorithmic steps such as color segmentation using OpenCV (HSV space), pixel equivalent calibration using a ruler, tracking the ball's center based on inter-frame difference, and finally, quadratic polynomial fitting of the displacement-time data to calculate the g-value. The code header explicitly declares adjustable parameters such as "starting frame offset" (range -50 to 50 frames), "rotation correction angle" (range -10 to 10 degrees), and "actual ruler width" (unit: meters).
[0050] The system runs this code to perform initial analysis on the uploaded video, obtaining an initial g value (e.g., 9.79 m / s²) and a goodness-of-fit R². The visualization results are as follows: Figure 5 As shown, it includes video tracking footage, displacement-time fitting curves, and real-time g-value display.
[0051] Subsequently, students accessed the interactive interface (such as...) Figure 6 The generated slider adjusts the "starting frame offset". Based on the dependency graph, the parametric analysis engine's computation scheduling unit identifies that this parameter only affects the "trajectory extraction" node. Therefore, it only re-executes the "trajectory extraction", "velocity calculation", and "g-value fitting" nodes, skipping upstream nodes such as "video decoding" and "color calibration". Traditional full-process recalculation takes approximately 2.1 seconds, while this system's selective recalculation only takes about 0.7 seconds, with a smooth interface response. Students can intuitively observe the impact of starting frame selection on measurement results and understand the sources of system errors. The data recording module automatically saves results under multiple sets of parameters and can generate a "starting frame offset - g-value" relationship curve for analysis.
[0052] Example 2: Investigation of Momentum Conservation in Collision Processes (Real Footage and Simulation Modes) This embodiment uses a "collision motion" experiment as a scenario to demonstrate how the system can uniformly process live video and physical simulation, and support in-depth parameter exploration.
[0053] The system supports two data sources: one is externally imported real-life collision videos (such as those on an air cushion track); the other is simulation videos generated in real time by the system's built-in physics engine based on user-defined parameters. The user input command is: "Create a collision experiment momentum analysis system, supporting real-life video analysis and simulation, allowing adjustment of the two spheres' mass, initial velocity, and coefficient of restitution, and real-time display of collision animation, momentum change curves, and energy loss rate." The code generation module generates dual-mode analysis code based on this. For the live-action mode, the code includes functions such as color threshold-based segmentation of the two spheres, contour center localization, and velocity calculation. For the simulation mode, the code embeds a collision physics model based on momentum conservation and the coefficient of restitution formula. The code declares general parameters such as "sphere 1 mass," "sphere 2 mass," and "coefficient of restitution," as well as "data source mode" switching parameters. This allows the same parameter system to control two different underlying processing pipelines.
[0054] In live-action mode, students upload videos, and the system automatically analyzes and plots a curve of momentum changing over time (e.g., ...). Figure 7 The total momentum line is approximately horizontal, providing a direct verification of the law of conservation of momentum.
[0055] In simulation mode, students can precisely set parameters for exploration. For example, with a fixed mass and initial velocity, only the "coefficient of restitution" can be adjusted from 0 (perfectly inelastic) to 1 (perfectly elastic). Based on parameter dependencies, the system only needs to re-calculate the velocity and energy loss rate after each adjustment, instantly updating the simulation animation and curves. The "coefficient of restitution - energy loss rate" data automatically recorded by the data recording module can be generated with one click, such as... Figure 4 The relationship curves shown clearly reveal the physical laws. By comparing real-world data with simulation results under different coefficients of restitution, students can verify the model and estimate the coefficient of restitution for actual collisions.
[0056] Example 3: Chaotic Motion Analysis of a Double Pendulum (Based on Physical Model Simulation) This example uses the "double pendulum chaotic system" from university physics as a scenario to demonstrate the system's ability to handle complex nonlinear dynamic simulation and analysis.
[0057] The user inputs the command: "Develop a tool for analyzing the chaotic motion of a double pendulum, which can adjust the length of the upper and lower pendulums and the initial angle, display the motion animation and phase space trajectory in real time, calculate the Lyapunov exponent, and analyze the chaotic characteristics." The code generation module matches the "Complex Dynamical System Simulation Template" and generates code that includes advanced algorithms such as the derivation of the Lagrange equation for the double pendulum, the Runge-Kutta method for numerical integration, and the calculation of the Lyapunov exponent based on the perturbation method. The code explicitly declares key parameters such as "upper pendulum length," "lower pendulum length," "initial angle of the upper pendulum," and "initial angle of the lower pendulum."
[0058] After students set a set of parameters (e.g., pendulum length is 1 meter, initial angle is 90 degrees), the system solves the equations of motion in real time, generates a simulation animation of the double pendulum motion, and simultaneously plots the phase space trajectory (θ-ω curve), calculates and displays the Lyapunov exponent (e.g., ...). Figure 8 The motion state is determined to be "weak chaos".
[0059] When students fine-tune the "initial swing angle" through the interactive interface, the computational scheduling unit recognizes this parameter as the system's initial condition. This triggers a recalculation of all motion states and chaotic exponents, starting from numerical integration, but does not affect irrelevant aspects such as graphics rendering. Students can immediately observe the phenomenon of the phase diagram transforming from a chaotic attractor to a regular periodic orbit, thus gaining a deep understanding of the extreme sensitivity of chaotic systems to initial conditions. The system records Lyapunov exponents under multiple sets of parameters, which can be used to plot the "initial angle-chaotic exponent" relationship graph and quantitatively analyze chaotic characteristics.
[0060] The above embodiments fully demonstrate the effectiveness of the system and the superiority of the method of the present invention. By rapidly customizing analysis tools using natural language, achieving flexible interaction through explicit parameters and automatically generated controls, ensuring real-time performance through dependency-driven selective recalculation, and integrating real-world and simulation analysis through a unified framework, the present invention provides a powerful new interactive analysis platform for physics experiment teaching and research. The above descriptions are merely embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.
Claims
1. An interactive physical experiment video analysis system based on generative artificial intelligence, characterized in that, include: The code generation module is used to receive a functional description in natural language and generate executable video analysis program code, wherein at least one adjustable analysis parameter is explicitly declared in the program code according to predefined parameter declaration rules; A parametric analysis engine, used to load and run the video analysis program code, includes: The parameter dependency resolution unit is used to establish a dependency graph between the adjustable analysis parameters and the specific analysis and calculation steps in the video analysis process when the program code is running. The computation scheduling unit is used to, upon receiving an update instruction for any of the adjustable analysis parameters, trigger recalculation only for analysis calculation steps that have a direct or indirect dependency relationship with the parameter, based on the dependency graph. The interactive interface module is used to dynamically parse the adjustable analysis parameters and their metadata declared in the program code at runtime, and automatically generate and present the corresponding graphical adjustment controls according to the data type of the parameters and the metadata constraints. The real-time visualization module is used to receive the updated analysis results output by the parameterized analysis engine after the value of the adjustable analysis parameter changes, and to perform real-time visualization rendering and output. The data recording and analysis module is used to automatically record different parameter settings and their corresponding analysis results, and supports the generation of correlation analysis charts between parameters and results based on the recorded data.
2. The interactive physical experiment video analysis system based on generative artificial intelligence according to claim 1, characterized in that, The code generation module is specifically used to: based on the natural language function description, call the large language model and combine it with the preset physical experiment analysis code template to generate the structured video analysis program code; the code template defines a general analysis process framework and parameterizable interface for different types of physical experiments.
3. The interactive physical experiment video analysis system based on generative artificial intelligence according to claim 1, characterized in that, The parametric analysis engine is configured to support unified processing of at least two data source types, including: a) Externally imported live-action physics experiment video data; b) Simulation video data is calculated and generated in real time by the system's built-in physical simulation engine based on the user-defined physical model parameters.
4. The interactive physical experiment video analysis system based on generative artificial intelligence according to claim 1, characterized in that, The adjustable analysis parameters include at least one of the following categories: a) Image processing parameters, including color threshold, filter kernel size, and contour detection sensitivity for target recognition; b) Kinematic analysis parameters, including the starting frame for trajectory extraction, sampling interval, and smoothing window size; c) Dynamic model parameters, including the object's mass, elastic restitution coefficient, damping coefficient, and gravitational acceleration.
5. The interactive physical experiment video analysis system based on generative artificial intelligence according to claim 1, characterized in that, When a recalculation is triggered, the computation scheduling unit automatically identifies and reuses all intermediate computation results in the current analysis process that have not been affected by the updated parameters.
6. The interactive physical experiment video analysis system based on generative artificial intelligence according to claim 1, characterized in that, The real-time visualization module is configured to output at least two visualization formats simultaneously, including: the original video footage with motion trajectory markers and physical quantity annotations superimposed, a physical quantity curve that changes over time, and a dynamic display panel of key calculation result values.
7. The interactive physical experiment video analysis system based on generative artificial intelligence according to claim 1, characterized in that, The graphical control generated by the interactive interface module includes one or more combinations of sliders, numeric input boxes, drop-down selection boxes, and switch buttons.
8. A method for analyzing interactive physical experiment videos based on generative artificial intelligence, characterized in that, Applied to the system as described in any one of claims 1-7, the method comprises: Responding to the functional description in natural language form, the code generation module generates video analysis program code containing explicit declarations of adjustable analysis parameters; The video analysis program code is loaded and run by the parametric analysis engine to perform analysis and calculation on the input video data. At the same time, the parameter dependency parsing unit establishes the dependency relationship between parameters and calculation steps, and the interactive interface module automatically generates parameter adjustment controls. The interactive interface module receives dynamic adjustment operations from the user on at least one of the adjustable analysis parameters. In response to the adjustment operation, the computation scheduling unit of the parameterized analysis engine selectively triggers the recalculation of related analysis and computation steps based on the dependency relationship, and drives the real-time visualization module to update the output. The data recording and analysis module stores the current parameter combination and the corresponding analysis results.
9. The interactive physical experiment video analysis method based on generative artificial intelligence according to claim 8, characterized in that, The "selective triggering of recalculation of associated analysis and computation steps" specifically includes: locating the affected node in the dependency graph based on the parameter update type, executing only the downstream computation chain starting from that node, and skipping computations related to upstream and other unrelated branches.
10. The interactive physical experiment video analysis method based on generative artificial intelligence according to claim 8, characterized in that, The method further includes: in the data recording and analysis module, comparing the analysis results under multiple parameter settings according to user selection, and automatically generating a curve or relationship chart reflecting the impact of parameter changes on the results.