Operational Safety

Operational Safety Features

AlzaWare combines real-time vision, role-based medical assistance, and cloud-backed intelligence to reduce risk, improve awareness, and support safer daily operation for Alzheimer patients.

Identity Safety Layer

Face recognition runs through MTCNN + FaceNet + SVM (RBF) to identify trusted people and reduce confusion in real-time interactions.

  • Training Accuracy: 99.12%
  • Validation Accuracy: 98.61%
  • Testing Accuracy: 97.22%
  • Low-confidence faces are filtered to avoid unsafe decisions

Medicine Safety Detection

A fine-tuned YOLOv8n model detects 12 medicine classes and daily objects, helping users avoid medication mistakes and improve situational awareness.

  • Precision: 0.9849
  • Recall: 0.9702
  • mAP@50: 0.995
  • Works in real-time on Raspberry Pi hardware

Clinical Support Chatbot

Gemini powers role-based guidance for Patient, Caregiver, and Doctor modes. Doctor mode integrates MRI-stage analysis with ResNet50 (80% accuracy) as a decision-support tool.

  • Multimodal inputs: voice, text, and image
  • RAG context from medical database + memory store
  • Supports safer and faster medical communication

Continuous Learning & Reliability

The system retrains safely when new faces are added: preprocessing, augmentation, embedding merge, then model refresh without service downtime.

  • 15 mobile images trigger retraining workflow
  • 8 augmented samples generated per uploaded image
  • >85% accuracy for newly added users
  • Hot model reload from AWS S3 to active inference service

Safety Operation Flow

1. Capture scene with camera
2. Route to face/object mode
3. Run AI inference pipeline
4. Validate confidence & context
5. Display safe guidance on OLED