Gaming Table Image Cropping for ML Resolution Limits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Casino gaming systems face challenges in efficiently tracking various objects and player interactions due to mismatched image quality requirements between modern cameras and neural network models, leading to slow game play and reduced revenue.
Innovation Solution
A gaming system dynamically generates and processes digital images to meet the maximum resolution target input requirement for machine learning models, allowing concurrent analysis and reducing network transmission time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the image resolution is increased to improve object identification accuracy, then the neural network model can identify objects with higher confidence, but the image file size increases and cannot be accepted by the neural network model due to maximum resolution limits
Solution Approach 1:
The patent divides the full-resolution image into multiple cropped regions of interest (ROIs) that are then processed separately. Each cropped image is resized to meet the neural network's maximum resolution requirement while maintaining sufficient detail for accurate object identification in that specific region.
Solution Approach 2:
The patent applies different processing quality levels to different regions of the image based on their importance. High-resolution processing is applied to regions where objects are detected, while lower-resolution processing is applied to empty or less critical regions, optimizing the balance between accuracy and model compatibility.
2Adaptability or versatility
If the image resolution is reduced to meet the neural network model's maximum input limit, then the image can be processed, but the pixel size becomes too small for the neural network model to recognize objects
Solution Approach 1:
Instead of uniformly reducing the entire image resolution, the patent segments the image into multiple cropped regions. Each region is then resized independently to the required resolution, ensuring that objects within each region maintain sufficient pixel density for recognition while the overall process complies with the model's input requirements.
Solution Approach 2:
The patent transitions from processing a single large image to processing multiple smaller images in a dimensional shift from one comprehensive view to multiple focused views. This allows the system to maintain high effective resolution for object detection while meeting the neural network's resolution constraints.
3Measurement precision
If multiple images are processed separately to meet neural network requirements, then each image can be analyzed, but the processing time increases and game play slows down
Solution Approach 1:
The patent merges multiple image processing operations into a single coordinated workflow. Multiple cropped images are processed in parallel and their results are combined to form the final object identification output, reducing total processing time compared to sequential processing while maintaining comprehensive detection coverage.
Solution Approach 2:
The patent performs preliminary actions by pre-identifying regions of interest and pre-cropping images before the main neural network analysis. This preliminary processing organizes the data structure and reduces the computational burden during the actual neural network inference, speeding up the overall process.
4Measurement precision
If high-resolution images are transmitted across the communications network to remote neural network models, then detailed object information can be analyzed, but the transfer time increases significantly
Solution Approach 1:
The patent extracts only the necessary regions of interest from the full-resolution image and transmits only these cropped portions to the remote neural network model. This extraction approach significantly reduces the amount of data transmitted over the network while preserving the essential object information needed for accurate analysis.
Solution Approach 2:
The patent segments the image transmission process into sending multiple small cropped images rather than one large image. This segmentation reduces the total data volume and transfer time while maintaining the ability to analyze objects in detail through the combined results of the cropped image analyses.
Data Source
AI summary
A gaming system that receives a frame of image data captured by a camera at a gaming table, detects, based on analysis of the image, a game state, and generates, via a graphical template associated with the game state, a set of digital images from cropped from portions of the frame of image data specified via the template. The system determines whether the set of images meets a maximum resolution target input requirement of a machine learning model. If the input requirement is met, the set of images are provided to the machine learning model as a unit for concurrent analysis. If the set of images does not meet the input requirement, the gaming system modifies, by an incremental amount, an image property of a subset from the set of images (e.g., reduces resolution of largest image in the set) until the set of images meets the input requirement.