Currently implemented features:
- Localisation of player on a reference map based on minimap
Mid-term planned features:
- Basic player stats (kills, assists, damage) monitoring using OCR
- Player state monitoring based on CV
- Game state monitoring based on CV
- Advanced player stats monitoring (knocks, times knocked, times respawned, time spent in loot boxes)
Long-term planned features:
- Deep learning on extracted stats for player improvement
Localisation algorithm:
- Crop current frame to contain only minimap
- Compute SIFT features for reference map and minimap
- Match features from reference map to minimap
- Using the distance ratio to determine the best matches
- Compute the scale difference between minimap and reference map by cross calculating all distances between the selected matched features
- Using the scale difference, estimate the ego position by projecting the minimap centre on the reference map
- Linear Kalman Filter to smooth position estimate