Dynamic Element Pose Determination with Temporal State Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pose determination methods struggle with low calculation efficiency and difficulty in simulating dynamic elements, particularly in real-time interactive scenarios such as games and virtual fitting, where elements like cloth and hair exhibit complex bending deformations.
Innovation Solution
A pose determination method that utilizes current object information and historical state information to determine the pose of elements in an image frame, employing a neural network-based approach with encoding and decoding modules, including timing modules and loss functions to improve efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional pose determination methods are used, then implementation is straightforward, but calculation efficiency is low and simulation of dynamic elements is difficult
Solution Approach 1:
The patent segments the pose determination process into distinct modules: an encoding module that processes current object information, a timing module that incorporates historical state information with time decay weighting, and a decoding module that generates final pose information. This modular segmentation enables parallel processing and optimizes calculation efficiency for dynamic element simulation.
Solution Approach 2:
The patent introduces an intermediate representation layer between input image data and output pose information. The encoding module transforms input data into intermediate features, which are then processed by the timing module to incorporate temporal dynamics, before being decoded into final pose information. This intermediary layer facilitates efficient computation while maintaining simulation accuracy.
2Manufacturing precision
If complex simulation effects are achieved, then visual fidelity is high, but CPU load increases
Solution Approach 1:
The patent changes the parameter representation from raw pixel data to compressed pose parameters. By determining pose information (such as joint positions, rotation angles, and skeletal structure) directly from image inputs through the neural network modules, the system achieves high-fidelity simulation effects while using fewer computational resources compared to processing full-resolution image data for each simulation calculation.
Solution Approach 2:
The timing module performs preliminary processing of historical state information by applying time decay weighting before feeding it to the decoding module. This preliminary action prepares the temporal data in an optimized format, reducing the computational burden during the main simulation calculation while maintaining the fidelity of dynamic element behavior over time.
3Speed
If real-time processing is implemented, then interaction responsiveness is improved, but accuracy of pose determination decreases
Solution Approach 1:
The timing module continuously integrates historical state information with current object information, applying a time decay function that weights recent states more heavily. This continuous integration maintains responsiveness for real-time processing while preserving temporal context that improves pose determination accuracy, as the system considers both current and historical states rather than isolated frames.
Solution Approach 2:
The system implements feedback by feeding historical pose information back into the timing module, which then influences the current pose determination. This feedback mechanism allows the system to maintain real-time processing speed while improving accuracy through temporal consistency, as the decoded pose information from previous frames informs the current state prediction and corrects for temporal variations.
Data Source
AI summary
Embodiments of the present disclosure provide a pose determination method and apparatus, an electronic device, and a storage medium. The method includes: obtaining current object information of a target object and historical state information of a target element, where the current object information is object information of the target object in a current image frame, the historical state information is state information of the target element in a historical image frame, and the target element is associated with the target object; and determining current pose information of the target element based on the current object information and the historical state information, where the current pose information is pose information of the target element in the current image frame.