👁️🤖 Real-Time Vision Intelligence (Prompt-Driven AI)

Can AI sense when it’s being watched?
This project demonstrates a real-time, prompt-driven Vision Intelligence system that doesn’t just detect objects — it understands scenes, intent, and behavior using Vision-Language Models.

---

🚀 Project Overview

This application connects a live camera feed to a Vision-Language Model (VLM) and allows users to dynamically change prompts to control what the AI analyzes — without retraining or restarting the system.

The same camera can instantly switch roles:

• From general scene understanding
• To behavior detection
• To safety monitoring
• To security surveillance

All controlled purely through natural-language prompts.

---

🎥 What Happens in the Demo

1. The AI first describes the live scene in front of the camera in natural language
2. The prompt is then changed in real time to:

   “Check if the person is using a smartphone”

3. The AI immediately adapts and correctly detects smartphone usage
4. No model retraining, no rule changes — just prompt-based intelligence

---

🔍 Why This Is Different

Traditional systems rely on fixed object detection and hard-coded rules.

This system enables:

• 🔄 Dynamic task switching
• 🧠 Contextual reasoning
• 📹 Per-camera custom intelligence
• 📝 Real-time text alerts

---

🧪 Example Prompt-Driven Use Cases

• 🦺 Alert if a worker is not wearing safety equipment
• 🔥 Detect fire, smoke, or sparks
• 📵 Detect mobile phone usage in restricted zones
• 🚧 Identify unsafe behavior near machinery
• 👥 Detect overcrowding or abnormal movement
• 🚫 Alert if someone enters a restricted area
• 🛡️ Identify suspicious perimeter activity

---

🏭 Potential Application Areas

• Industrial safety & compliance
• Manufacturing & warehouses
• Smart surveillance
• Defense & perimeter security
• Critical infrastructure
• Airports & high-security zones

---

🧠 Technology Stack

Hardware

• NVIDIA RTX 3050 (6GB VRAM)

AI Models

• Gemma 3
• Qwen2.5-VL
• Served locally via Ollama

Software

• Python
• OpenCV
• PyQt6

---

⚙️ How to Run

python vision-app.py

Ensure:

• Webcam is connected
• Ollama is running locally

---

💡 Key Takeaway

Vision + Language + Prompts = Adaptive Intelligence

---

🤝 Open to

• Collaborations
• Real-world deployments
• AI / Computer Vision roles

---

Screenshots

🔖 Tags

#AI #ComputerVision #EdgeAI #IndustrialAI #DefenseTech #Python #NVIDIA #RTX #Ollama