ROI Upscaling for OCR-Based Game Event Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting events in video game content, especially without access to game code, require significant training data and resource-intensive model training, leading to inefficiencies in content presentation and increased manual intervention.
Innovation Solution
A system utilizing detectors to analyze specific regions of the game display, such as HUD elements, for pattern recognition, combined with machine learning models and event translation modules, to identify events without complex training, and customizable for different games.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional event detection methods are used without access to game code, then event detection can be performed, but significant training data and resource-intensive model training are required
Solution Approach 1:
The patent extracts and analyzes specific regions of interest (such as HUD elements, health bars, ammo counters) from the game display without requiring access to the entire game code or comprehensive training data. By focusing detection efforts on these specific extracted regions, the system achieves reliable event detection with reduced complexity
Solution Approach 2:
The game display is segmented into multiple regions of interest, each potentially containing different types of event information. The system processes these segmented regions independently using detectors, which reduces the overall computational complexity and training data requirements compared to analyzing the entire display as a single unit
2Reliability
If traditional event detection methods are used, then event detection can be performed, but manual intervention is required to create content of interest
Solution Approach 1:
The system automatically detects events and generates content without requiring manual intervention. The detectors automatically identify events of interest in the game display, and the event translation module automatically translates these detections into structured event data, enabling fully automated content creation from gameplay
Solution Approach 2:
The system implements a feedback loop where detected events are translated into structured data that can be used to automatically generate and present content. This automated feedback mechanism eliminates the need for manual content creation while maintaining high reliability in event detection
3Productivity
If manual intervention is used to create content, then content can be presented to viewers, but the time before content can be presented is increased
Solution Approach 1:
The system performs preliminary detection and translation of events as they occur during gameplay, preparing content for immediate presentation. By detecting and translating events in real-time rather than requiring post-processing, the system eliminates delays in content presentation
4Reliability
If resource-intensive model training is used, then event detection can be performed, but resource consumption is high
Solution Approach 1:
Instead of applying complex trained models to the entire game display, the system applies simpler detectors to specific local regions of interest. Each detector is tailored to its specific region (e.g., health bar detector for the health bar area, ammo detector for the ammo counter area), achieving high detection accuracy with reduced computational resources
Data Source
AI summary
A game-agnostic event detector can be used to automatically identify game events. Game-specific configuration data can be used to specify types of pre-processing to be performed on media for a game session, as well as types of detectors to be used to detect events for the game. Event data for detected events can be written to an event log in a form that is both human- and process-readable. The event data can be used for various purposes, such as to generate highlight videos or provide player performance feedback. The event data may be determined based upon output from detectors such as optical character recognition (OCR) engines, and the regions may be upscaled and binarized before OCR processing.