Image Processing Apparatus for Player Action Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies for recognizing player actions in sporting events using shot images of entire courts struggle with low-resolution images and inability to specify the player of interest without prior position determination, leading to inadequate recognition results.
Innovation Solution
An image processing apparatus that includes obtaining, detecting, trimming, and recognizing player actions by obtaining images with the court in view, detecting multiple objects, trimming specific player areas based on object positions, and recognizing actions using trimmed images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If a shot image of the entire court is used to detect players, then the position of players can be detected, but the resolution becomes too low to recognize face features and movements accurately
Solution Approach 1:
The patent segments the full court image into multiple smaller images based on detected player positions. Each small image focuses on a specific player or region, allowing high-resolution analysis of face features and movements while maintaining coverage of the entire court through the collection of segmented images.
Solution Approach 2:
The patent extracts relevant regions containing players from the full court image. By identifying player positions and extracting only those specific regions for detailed analysis, the system obtains high-resolution data for recognition while avoiding the need to process the entire low-resolution court image.
2Productivity
If machine learning is applied to low-resolution court images, then processing load is reduced, but sufficient pixel information for recognizing face and movement is not obtained
Solution Approach 1:
The patent performs preliminary processing to detect player positions and extract relevant regions before applying machine learning. This preliminary action identifies which regions contain players and extracts only those regions, ensuring that the machine learning process receives high-resolution input data while maintaining processing efficiency through targeted analysis.
Solution Approach 2:
The patent extracts high-resolution player regions from the full court image and uses only these extracted regions for machine learning analysis. This extraction ensures sufficient pixel information is available for recognition while keeping the overall processing load manageable by analyzing only the necessary regions rather than the entire image.
3Ease of operation
If the position of players is determined in advance to specify the player of interest, then the player can be identified, but this determination cannot be made for all players in the shot image without prior position knowledge
Solution Approach 1:
The patent enables the system to automatically determine player positions and identify players of interest without requiring external position information. The image processing apparatus detects player positions, extracts relevant regions, and identifies the player whose action needs recognition based on the extracted image content itself, making the system self-sufficient and eliminating the need for pre-provided position data.
Solution Approach 2:
The patent performs preliminary detection of player positions and extraction of player regions before identifying the specific player of interest. This preliminary action provides the necessary position information automatically, enabling the system to specify any player in the shot image without requiring external position determination or prior knowledge about player locations.
Data Source
AI summary
An image processing apparatus obtains an image shot such that a court of a sporting event is within an angle of view, detects a plurality of objects in the image, trims an area specified based on positions of the plurality of objects in the image, and recognizes an action of a specific player out of the plurality of objects based on a trimmed image. The apparatus trims an area including the specific player specified based on a reference being a position of a first object used in the sporting event out of the plurality of objects.