BISINDO Real-Time Detection

Developed a real-time BISINDO alphabet detection system using the YOLOv8 object detection architecture to recognize hand gestures directly from live camera input.

Most BISINDO recognition research focuses on offline evaluation rather than real-time interaction, limiting its practicality for everyday communication.

Built a real-time detection pipeline using YOLOv8 capable of recognizing BISINDO alphabet gestures directly from webcam input with low-latency inference.

• Selected YOLOv8 to achieve real-time inference performance.
• Built a custom training dataset tailored for BISINDO alphabet detection.
• Focused on balancing inference speed with recognition accuracy.

• Maintaining stable detection under varying lighting conditions.
• Handling gesture similarity between multiple BISINDO characters.
• Collecting sufficiently diverse training samples.

• Real-time deployment introduces challenges beyond offline evaluation.
• Dataset diversity significantly influences detection robustness.
• Inference speed is equally important as model accuracy for interactive applications.

• Sentence-level BISINDO recognition.
• Temporal gesture modeling.
• MediaPipe-assisted hand tracking.
• Mobile deployment.