🔗 Interactive Dashboard:
https://fardinsk25.github.io/SmartAssist_Vision_Data_Analysis/
SmartAssist Vision is an AI-powered wearable mobility aid designed to assist visually impaired individuals through real-time object detection, obstacle awareness, distance estimation, and voice-guided navigation.
The system combines:
- MobileNet-SSD Object Detection
- Raspberry Pi 4 Model B
- HC-SR04 Ultrasonic Sensor
- Voice Navigation using eSpeak / pyttsx3
This repository focuses on the performance evaluation and analytics of the prototype using a simulated evaluation dataset containing 5,000 detection records.
The project demonstrates how data analytics can be used to evaluate AI system performance through detection accuracy, confidence analysis, response time monitoring, safety metrics, and operational insights.
| Metric | Value |
|---|---|
| Total Detection Records | 5,000 |
| Detection Accuracy | 92.56% |
| Average AI Confidence | 0.926 |
| Average Response Time | 145.9 ms |
| High-Risk Alerts | 1,373 |
| Average Battery Level | 73.9% |
| Very Close Detections | 1,500 |
| Night Evaluations | 1,483 |
| Prototype Device | Raspberry Pi 4 + MobileNet-SSD |
Wearable prototype featuring Raspberry Pi 4, USB camera, ultrasonic sensor, and earphone-based voice guidance.
- KPI Monitoring
- Detection Accuracy Tracking
- Monthly Performance Trends
- Response Time Analysis
- Battery Monitoring
- Object Detection Distribution
- Detection Status Analysis
- Position Classification
- Distance Category Breakdown
- Environment-Based Performance
- Confidence Distribution
- Confidence by Object Class
- Monthly Confidence Trends
- Response Time Stability
- Model Reliability Analysis
- Scenario-Based Evaluation
- Detection Accuracy Comparison
- Battery Performance Analysis
- Response Time Tracking
- Performance Matrix
- Accuracy improved from 91.6% to 94.4% across the evaluation period.
- Daily accuracy remained consistently above 85%.
- Person was the most frequently detected object.
- Average confidence score reached 0.926.
- More than 55% of detections exceeded 0.95 confidence.
- Day and night performance remained highly consistent.
- 30% of detections were classified as Very Close (<50 cm).
- Person and Dog generated the highest number of High-Risk alerts.
- Distance-based alert generation remained stable throughout evaluation.
- Average response time was 145.9 ms.
- Performance remained comfortably below the 200 ms target threshold.
- Response times were consistent across all scenario groups.
- Python
- Pandas
- Matplotlib
- Seaborn
- Jupyter Notebook
- HTML
- CSS
- JavaScript
- Chart.js
- MobileNet-SSD
- OpenCV
- Raspberry Pi 4
- HC-SR04 Ultrasonic Sensor
- eSpeak / pyttsx3
SmartAssist_Vision_Data_Analysis
│
├── assets/
├── data/
├── notebook/
├── scripts/
│
├── index.html
├── README.md
├── requirements.txt
├── .gitignore
└── LICENSE
Visit:
https://fardinsk25.github.io/SmartAssist_Vision_Data_Analysis/
Clone the repository:
git clone https://github.com/YOUR_USERNAME/SmartAssist_Vision_Data_Analysis.gitInstall dependencies:
pip install -r requirements.txtOpen:
index.html
in any modern browser.
- YOLOv8 Integration
- Edge TPU Acceleration
- GPS-Assisted Navigation
- Mobile Application Integration
- Real-Time Cloud Monitoring
- Live Performance Analytics
This project is licensed under the MIT License.
Data Analytics • AI • IoT • Computer Vision
🌐 Portfolio: https://fardinsk25.github.io/portfolio/
💼 LinkedIn: https://www.linkedin.com/in/fardinshaikh02/
🔗 Live Dashboard:
https://fardinsk25.github.io/SmartAssist_Vision_Data_Analysis/
🐙 GitHub:
https://github.com/fardinsk25
⭐ If you found this project interesting, consider giving the repository a star.