VISTA — Your Intelligent Safety Guardian

VISTA doesn't just monitor your car — it protects you and keeps your loved ones informed. The system continuously learns from sensor data, camera vision, and driving patterns to predict problems before they become expensive repairs or safety hazards.

---

🎥 VISTA Explainer & Live Demonstration

⚡ Click the player above to watch the VISTA HD explainer video! ⚡
_{(Visualizing edge AI collision alerts, YAMNet acoustic processing, real-time CAN/OBD telemetry, and Gemini-powered damage analytics)}

---

📑 Table of Contents

• What You Get
• Enterprise Value
  • For Insurance Companies
  • For Automotive Manufacturers
• Hardware & BOM
• Engineering & Architecture
• Key Design Decisions
• System Verification
• Quick Start
• Documentation & Project Structure
• Team & License

---

⚡ What You Get

• 🚨 Instant crash alerts to you and emergency contacts with precise location data.
• 🔒 Theft detection & real-time notifications the moment unusual activity occurs.
• 🔧 Predictive maintenance warnings that catch issues before breakdowns happen.
• 📊 Driving pattern analysis that identifies risky behaviors and suggests improvements.
• 🛠️ Pre-tuned maintenance schedules shared directly with your trusted mechanic.

The longer VISTA runs, the smarter it becomes — learning your vehicle's unique behavior to separate normal wear from genuine concerns. It's not just protecting your car; it's protecting you.

---

🏢 Enterprise Value

🛡️ For Insurance Companies: The Truth Layer

Insurance fraud costs the industry billions annually. VISTA provides an intelligent black box that doesn't just record — it analyzes, contextualizes, and delivers forensic-grade incident data.

• Undeniable incident reconstruction with multi-sensor data fusion (vision + telemetry)
• Fraud detection patterns identified through AI analysis of crash signatures
• Pre-crash behavior analysis showing driver actions in the critical seconds before impact
• Automated claims validation reducing investigation time from weeks to minutes

Replace guesswork with ground truth. VISTA transforms claims processing from adversarial negotiation into objective data analysis.

🏭 For Automotive Manufacturers (NVH): Global Fleet Intelligence

Traditional R&D costs billions and relies on controlled test environments. VISTA delivers something far more valuable: real-world, structured intelligence from thousands of vehicles operating in actual conditions.

• Component failure patterns identified across your entire fleet before recalls become necessary
• Real-world performance metrics from diverse geographies, climates, and driving conditions
• Early warning system for design flaws that only emerge in production environments
• Competitive intelligence through aggregate performance benchmarking

VISTA turns every vehicle into a distributed R&D sensor, potentially saving hundreds of millions in traditional testing costs.

🔗 The Common Thread

Each stakeholder receives fundamentally different value from the same underlying platform. VISTA adapts its intelligence layer to serve the unique needs of drivers, insurers, and manufacturers — proving that in the modern era, the same data can unlock entirely different forms of value.

---

🔌 Hardware

Total BOM: ₹5,770 (~$69 USD) · Recurring cloud cost: ₹0/month

Component	Purpose	Cost
Raspberry Pi 4B (4GB)	Main compute	₹0 (owned)
ESP32-C3-DevKitM-1	Always-on 5μA sentinel	₹400
ELM327 OBD-II (USB)	Vehicle CAN data	₹500
MPU6050 GY-521	Crash + motion detection	₹150
Pi Camera v3 (IMX708)	Scene capture for Gemini	₹1,800
USB Microphone	YAMNet audio classification	₹200
HC-SR501 PIR	Parked intrusion detection	₹80
AO3401 P-MOSFET + 2N2222	Pi power control circuit	₹60
Fuel pump relay module	Ghost Key physical immobilizer	₹60
DC-DC LM2596 (12V→5V)	Car power regulation	₹300
Kingston A400 SSD (120GB)	Event DB + images (not SD)	₹900
Misc (wires, resistors, enclosure)	Assembly	₹320

The USB SSD is mandatory, not optional. SQLite in WAL mode and InfluxDB write-heavy workloads destroy SD card flash cells within weeks.

See → raw/DESIGN/DESIGN_v4/docs/02_HARDWARE_DESIGN.md for complete pin diagrams, wiring schematics, and installation guide.

🛠️ Custom Hardware Stack & PCB Layouts

VISTA's electrical engineering incorporates a custom ESP32-C3 coprocessor architecture, transient-protected DC-DC regulators, and active power MOSFET controls. Below are the design assets:

---

⚙️ Engineering & Architecture

VISTA is a vehicle-mounted intelligence system built on ₹5,770 of hardware, fusing four physical sensing modalities — OBD-II vehicle bus data, inertial measurement, acoustic classification, and camera — to detect safety-critical events and respond autonomously.

┌──────────────────────────────────────────────────────────────────┐
│                    VEHICLE SENSING LAYER                          │
│  OBD-II ELM327    MPU6050 IMU     USB Microphone   Pi Camera v3 │
│  /dev/ttyUSB0     I2C 0x68        16kHz mono        CSI-2 3MP   │
│  2 Hz actual      100 Hz          YAMNet window      On-demand   │
└─────────────────────────────┬────────────────────────────────────┘
                              │
┌─────────────────────────────▼────────────────────────────────────┐
│                HARDWARE ABSTRACTION LAYER (HAL)                    │
│  6 drivers with graceful demo-mode fallback                       │
│  System never crashes due to missing hardware                     │
└─────────────────────────────┬────────────────────────────────────┘
                              │
┌─────────────────────────────▼────────────────────────────────────┐
│                   INTELLIGENCE PIPELINE                            │
│                                                                    │
│  VelocityEKF          2-state Kalman: OBD + IMU fusion            │
│  CrashDetector        4-tier: jerk → audio → OBD → vision        │
│  AudioClassifier      YAMNet TFLite 4.1MB, 521 → 6 classes       │
│  TheftDetector        Ghost Key TSA: 4-layer temporal analysis    │
│  PredictiveAnalytics  NVH FFT + Gemini mechanic report            │
│  SystemHealthMonitor  Sensor liveness + CPU/RAM/temp              │
│  DecisionEngine       Weighted confidence, explainable output     │
│  CloudVision          Gemini 1.5-flash REST, 3 retries            │
└─────────────────────────────┬────────────────────────────────────┘
                              │
              ┌───────────────┴───────────────┐
              ▼                               ▼
┌─────────────────────┐         ┌─────────────────────────┐
│   COMMUNICATIONS    │         │   DATA STORAGE          │
│  Telegram (alerts)  │         │  SQLite (events, WAL)   │
│  MQTT (telemetry)   │         │  InfluxDB (time-series) │
│  BLE (proximity)    │         │  USB SSD primary        │
│  Buzzer (local)     │         │  (not SD card)          │
└─────────────────────┘         └─────────────────────────┘

┌──────────────────────────────────────────────────────────────────┐
│              ESP32-C3 COPROCESSOR (ALWAYS-ON)                    │
│  Deep sleep: 5μA · PIR intrusion · BLE auth · MOSFET Pi power   │
│  Battery monitor · Heartbeat watchdog · 4-state machine          │
└──────────────────────────────────────────────────────────────────┘

#### 🗺️ Unified System Block Diagram

The complete hardware, signal processing, edge logic, and secure BLE/cellular cloud communications topology:

<p align="center">
  <img src="assets/vista_system_block_diagram.png" width="95%" alt="VISTA System Architecture Block Diagram" style="border-radius: 8px;">
</p>

---

🔑 Key Design Decisions

Tiered sensor architecture, not equal weighting. The IMU responds in <10ms. Audio responds in ~50ms. OBD responds in 300–500ms. These sensors cannot be treated as equals. VISTA assigns them roles: IMU is the primary trigger, Audio is the fast corroborator, OBD is the async post-event confirmation.

Hardware-layer theft response. A physical relay on the fuel pump power wire cannot be defeated by CAN-bus injection or any software attack. This is the architectural reason for choosing an analog relay over a CAN command.

Persist before transmit. Crash events are written to SQLite before any Telegram alert is attempted. If the network fails, the event survives. This reversal of the naive "alert first" pattern is critical for forensic completeness.

The cloud does what the edge cannot. Gemini 1.5-flash replaces 5 separate local vision models (object detection, scene classification, damage assessment, license plate reading, natural language reporting). One API call. Zero local compute overhead. Context-aware output in the owner's language.

Every hardware claim must survive physics. OBD-II achieves 2–3 Hz in practice, not 10 Hz. Pi 4 cannot suspend-to-RAM. MPU6050 saturates at ±16g — deliberately configured at maximum range because real crashes produce 20–70g. These are not assumptions; they are measured facts baked into config.yaml.

⚡ Edge-Level Sensor Fusion

The asynchronous decision cascade prioritizing immediate IMU/jerk detection, fast YAMNet acoustic confirmation, post-event OBD-II telemetry verification, and cloud vision corroboration:

🔒 Ghost Key Anti-Theft Verification

Hardware-layer fuel pump immobilizer relay state flow, requiring active Bluetooth Low Energy (BLE) secure token handshake to authenticate:

---

🧪 System Verification

(Last verified: May 16, 2026)

python -m pytest src/vista/tests/test_v3_quick.py
→ 20 passed, 0 failed

python scripts/demo_billion_dollar_architecture.py
→ theft_detected:    ✅ PASS
→ legitimate_passes: ✅ PASS
→ nvh_pipeline:      ✅ PASS
→ ALL SCENARIOS PASSED. Exit code: 0

Import check:
→ 19/19 modules importable (DEMO_MODE=true)

🚨 Crash/Theft Forensic Analysis & Unit Placement

VISTA uses physical placement optimized for cabin soundscapes and vehicle axis alignment, combined with real-world Telegram visual warning cards generated in under 3.5 seconds:

---

🚀 Quick Start

git clone https://github.com/AdityaPagare619/VISTA.git
cd VISTA

# Configure API keys
cp src/vista/.env.example src/vista/.env
# Edit .env: add GEMINI_API_KEY, TELEGRAM_BOT_TOKEN, TELEGRAM_CHAT_ID

# Run in demo mode (no hardware needed)
cd src/vista
DEMO_MODE=true python run_dashboard.py
# Open http://localhost:5000

# Run on Raspberry Pi (production)
sudo ./scripts/deploy.sh
sudo systemctl start vista vista-dashboard

---

📚 Documentation

Document	For
DESIGN_v4/docs/README.md	Start here — index + reading order by team
01 System Design	Architecture, state machine, boot sequence, innovation claims
02 Hardware Design	BOM, wiring diagrams, GPIO tables, relay circuit
03 Software Architecture	Package structure, module APIs, data schemas
04 Operational Flows	Crash timeline, Ghost Key sequence, NVH flow, EKF dropout
05 Technology Stack	Dependencies, cloud config, deployment, CI/CD
06 Demo Methodology	Dashboard walkthrough, exam scripts, verifiable claims

---

📂 Project Structure

VISO-PROJECT/
├── README.md
├── pyproject.toml              ← pytest config
├── scripts/
│   ├── deploy.sh               ← One-command Pi deployment
│   └── demo_billion_dollar_architecture.py  ← SITL demo
├── .github/workflows/ci.yml    ← 4-job CI pipeline
└── src/vista/
    ├── main.py                 ← 5-phase initialization
    ├── config.yaml             ← All configuration
    ├── hal/                    ← 6 hardware drivers
    ├── intelligence/           ← 8 AI/ML modules
    ├── communication/          ← Telegram, MQTT, BLE
    ├── data/                   ← SQLite + InfluxDB
    ├── dashboard/              ← Flask-SocketIO web UI
    ├── esp32/main/main.c       ← 1,098 LOC C firmware
    ├── models/yamnet.tflite    ← 4.1MB real ML model
    └── tests/                  ← 20-test verified suite

---

👥 Team

Role	Domain
Hardware Lead	Wiring, sensors, ESP32 firmware, relay circuit, vehicle installation
AI/ML Engineer	YAMNet training, Gemini integration, NVH baseline collection
Data & Integration	Dashboard, alerts, database, system integration, CI/CD
Research & Docs	IEEE paper, documentation, presentation, evaluation

---

📜 License & Contribution

This project is licensed under the MIT License — use, modify, and build upon freely. Credit appreciated.

Contributions are welcome! Please feel free to submit a Pull Request.