Temporal 2D-3D Pose Estimation for Golf Swing Hosel Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current human pose estimation methods for golf swing analysis face challenges in accuracy due to fast-moving and partially self-occluded key points, particularly the golf club hosel, which is crucial for analyzing golf players' swings, and require significant computational resources, making them inefficient for mobile devices.
Innovation Solution
A lightweight temporal-based 2D and 3D human pose estimation method using temporal information, which removes one AI convolution layer to reduce computational resources while maintaining performance, incorporates traditional computer vision techniques for geometric constraints, and employs a line segment detection algorithm to improve golf club hosel prediction accuracy, merging deep learning and traditional computer vision results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If standard AI convolution layers are used for pose estimation, then prediction accuracy is improved, but computational resources increase significantly
Solution Approach 1:
The patent removes at least one AI convolution layer from the standard convolution cycles of the pose estimation model. This extraction of the computationally intensive convolution layer directly reduces the computational resources and energy consumption while maintaining acceptable prediction accuracy for pose estimation in golf swing analysis.
2Measurement precision
If more AI convolution layers are used, then prediction accuracy is improved, but device complexity increases
Solution Approach 1:
The patent simplifies the model architecture by removing at least one AI convolution layer from the standard convolution cycles. This reduction in model complexity makes the system more suitable for deployment on mobile devices with limited computational resources while maintaining adequate prediction accuracy for golf swing analysis.
3Use of energy by moving object
If traditional computer vision techniques are used alone, then computational resources are reduced, but accuracy for fast-moving and occluded key points deteriorates
Solution Approach 1:
The patent merges deep learning-based pose estimation with traditional computer vision techniques, specifically incorporating a line segment detection algorithm for golf club hosel prediction. This combination allows the system to leverage the strengths of both approaches: deep learning for general pose estimation and traditional computer vision for accurately detecting fast-moving and occluded key points like the golf club hosel, while keeping computational resources manageable.
Data Source
AI summary
Systems and methods are disclosed for computer vision and AI technology for implementing a temporal-based two-dimensional (2D) human pose estimation method for golf swing analysis using temporal information to improve the accuracy of fast-moving and partially self-occluded key points. The system may, for example, determine a bounding box to incorporate with an image received from a user device; initiate a 2D inference process on the image to generate a final 2D image; provide a final 2D image and set of confidence scores for each corresponding with the key point of the image to a three-dimensional (3D) inference process and Perspective-n-Point (PnP) process; using output from the 3D inference process and the PNP process, generate a 3D image that is altered in accordance with a distance value between the user device and the camera; and provide the 3D image to the user device.