BISINDO Sign Language Detection is a thesis project focused on building an instance segmentation model to detect and recognize Indonesian Sign Language (BISINDO) alphabet gestures in real-time.
The project implements YOLOv8 instance segmentation with a complete ML pipeline: data collection, annotation using Roboflow, data cleaning, and various augmentation strategies to improve model robustness.
A key contribution of the research is the evaluation of model performance across different data split scenarios and augmentation combination strategies, providing insights into optimal training configurations for sign language detection tasks.
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ COLLECT │────▶│ ANNOTATE │────▶│ AUGMENT │
│ Data │ │ Roboflow │ │ Strategies │
└──────────────┘ └──────────────┘ └──────┬───────┘
│
▼
┌──────────────┐
│ TRAIN YOLO │
│ v8-seg │
└──────┬───────┘
│
▼
┌──────────────┐
│ EVALUATE │
│ Cross-Split │
└──────────────┘Data Collection → Annotation → Augmentation → Training → Evaluation