Learning-Based Ground Position Estimation for Sports Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for determining the shooting position of an object in sports, such as basketball or netball, are often inaccurate due to obstructed views, inaccurate player detection, assumptions about the shooter's position, and the need for calibration, leading to computational intensity and reduced accuracy.
Innovation Solution
A learning-based system that uses a neural network to estimate the ground position of an object's release point by analyzing a series of images from a camera, eliminating the need for precise camera positioning and reducing computational requirements, while compensating for camera and target region movement and variations in player height or shot type.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods are used to determine shooting position, then the system can provide basic position estimation, but the accuracy is poor due to obstructed views and inaccurate player detection
Solution Approach 1:
The patent replaces conventional mechanical/computational methods for trajectory tracking and player detection with a learning-based system using neural networks. The neural network directly estimates ground position from image data, substituting complex mechanical tracking systems and achieving higher accuracy without requiring precise camera positioning or calibration.
Solution Approach 2:
The patent changes the approach from tracking multiple parameters (trajectory, player position, camera calibration) to a learning-based method that estimates ground position directly from image features. This parameter transformation allows the system to achieve accurate shooting position estimation without relying on precise measurement of intermediate parameters.
2Measurement precision
If conventional trajectory analysis methods are used, then the system can track object motion, but computational requirements increase due to the need for precise camera positioning and calibration
Solution Approach 1:
The patent replaces computationally intensive trajectory analysis and calibration procedures with a learning-based estimation system. The neural network performs ground position estimation directly from image data, eliminating the need for complex computational geometry calculations and precise camera positioning, thereby reducing computational energy requirements.
Solution Approach 2:
The patent employs a pre-trained neural network that has already learned the mapping from image features to ground position during the training phase. During operation, the system only needs to perform forward propagation through the trained network, which is computationally efficient compared to real-time trajectory analysis and calibration procedures.
3Measurement precision
If conventional player detection methods are used, then the system can identify shooter location, but accuracy is reduced due to assumptions about shooter position and obstructed views
Solution Approach 1:
The patent replaces conventional player detection and trajectory intersection methods with a learning-based system that directly estimates ground position from image data. This approach does not require detecting the player or making assumptions about shooter position, and is robust to occlusions since it learns to infer position from available visual features.
Solution Approach 2:
The patent introduces a neural network as an intermediary that maps image features directly to ground position estimation, bypassing the need for intermediate steps such as player detection, trajectory tracking, and geometric calculations. This intermediary learns to handle occlusions and variations in shooting scenarios during training.
Data Source
AI summary
Operations of the present disclosure may include receiving a group of images taken by a camera over time in an environment. The operations may also include identifying a first position of an object in a target region of the environment in a first image of the group of images and identifying a second position of the object in a second image of the group of images. Additionally, the operations may include determining an estimated trajectory of the object based on the first position of the object and the second position of the object. The operations may further include, based on the estimated trajectory, estimating a ground position in the environment associated with a starting point of the estimated trajectory of the object. Additionally, the operations may include providing the ground position associated with the starting point of the estimated trajectory of the object for display in a graphical user interface.