League of Legends Minimap Detection

Project Proposal: Champion Icon Detection and Analysis in League of Legends Mini-Maps

1. Main Project Idea or Theme

The primary objective of this project is to develop a computer vision model that accurately detects and delineates the icons of champions on the mini-map in League of Legends (LoL) gameplay videos. By drawing precise borders around these icons and determining the confidence levels of each detection, the model aims to provide detailed spatial and statistical insights into champion positions and movements throughout a game. This tool can serve as a foundational component for advanced analytics in competitive gaming, offering deeper understanding and new metrics beyond traditional in-game statistics.

2. Proposed Approach

To achieve accurate detection of champion icons on the mini-map, the project will follow a multi-step approach:

1. Data Preprocessing:
   • Extract frames from raw VOD (Video on Demand) footage at a consistent frame rate.
   • Normalize and resize frames to a standard resolution to ensure uniformity in input data.
2. Model Selection and Architecture:
   • Utilize a state-of-the-art object detection framework such as YOLO (You Only Look Once) or Faster R-CNN, known for their real-time detection capabilities and high accuracy.
   • Fine-tune a pre-trained model on a dataset specific to LoL mini-maps to leverage transfer learning and reduce training time.
3. Annotation and Labeling:
   • Create a labeled dataset by manually annotating champion icons on a diverse set of mini-map images to train the model effectively.
4. Training and Validation:
   • Split the dataset into training, validation, and testing subsets.
   • Implement data augmentation techniques to enhance model robustness against varying in-game conditions and visual noise.
5. Post-Processing:
   • Apply non-maximum suppression to eliminate redundant bounding boxes.
   • Calculate confidence scores for each detected icon to assess detection reliability.
6. Integration and Visualization:
   • Develop a user interface to visualize the detected icons and their confidence levels in real-time or from recorded gameplay videos.

3. Input/Output Data

• Input Data:
  • Raw VOD Footage: Video recordings of League of Legends gameplay, which are widely available across platforms like Twitch, YouTube, and official LoL streaming services.
  • Frame Extraction: Individual frames extracted from the VODs will serve as the primary input for the model.
• Output Data:
  • Bounding Boxes: Precise coordinates outlining each detected champion icon on the mini-map.
  • Confidence Scores: Numerical values representing the model's confidence in each detection, aiding in the assessment of detection reliability.

4. Training Data — Sources and Acquisition

To train the model effectively, a substantial and diverse dataset is essential. The following sources and methods will be employed to gather and prepare the training data:

1. VOD Collection:
   • Scrape and download VODs from popular streaming platforms like Twitch and YouTube, focusing on high-quality recordings to ensure clarity of mini-map visuals.
2. Manual Annotation:
   • Utilize annotation tools such as LabelImg or CVAT to manually label champion icons on mini-map images.
   • Ensure annotations cover various in-game scenarios, including different map phases, champion abilities that might obscure icons, and diverse team compositions.
3. Data Augmentation:
   • Apply transformations like rotation, scaling, and brightness adjustments to increase dataset variability and improve model generalization.
4. Dataset Diversity:
   • Include VODs from different regions, skill levels, and game modes to capture a wide range of mini-map presentations and champion icon variations.

5. Evaluation Plan

To assess the performance and effectiveness of the developed model, the following evaluation strategies will be implemented:

1. Quantitative Metrics:
   • Precision and Recall: Measure the model's ability to correctly identify champion icons (precision) and its capability to detect all relevant icons (recall).
   • Mean Average Precision (mAP): Evaluate the overall accuracy of the model across different IoU (Intersection over Union) thresholds.
   • F1-Score: Provide a balanced metric combining precision and recall to assess the model's accuracy.
2. Qualitative Analysis:
   • Conduct visual inspections of detection results on a set of unseen gameplay videos to identify and analyze any misdetections or missed icons.
   • Gather feedback from League of Legends players and analysts to understand the practical utility and areas for improvement.
3. Benchmarking:
   • Compare the model's performance against existing object detection models or any baseline methods to determine relative effectiveness.
4. Real-Time Performance:
   • Assess the model's inference speed to ensure it meets real-time processing requirements, which is crucial for live analytics applications.

6. Impact

While the project's focus is niche, its successful implementation holds significant value within the competitive gaming community and beyond:

1. Enhanced Player Analytics:
   • Provide detailed spatial data on champion positions and movements, enabling players and coaches to analyze strategies, positioning, and decision-making processes more effectively.
2. Pro Play Scoring:
   • Contribute to the development of advanced metrics for evaluating player performance in professional matches, potentially influencing player rankings and scouting.
3. Statistical Insights:
   • Generate novel statistics that are not captured by traditional in-game metrics, offering deeper insights into game dynamics and champion interactions.
4. Broadening Research Horizons:
   • Serve as a foundation for further research in esports analytics, encouraging the integration of computer vision techniques in other aspects of game analysis.
5. Community Engagement:
   • Foster a more data-driven approach within the League of Legends community, promoting informed discussions and strategic improvements among players.

In essence, this project bridges the gap between raw gameplay data and actionable insights, empowering stakeholders in the esports ecosystem with tools to enhance performance, strategy, and overall understanding of the game.

Index

winters...