A smart surveillance system powered by machine learning that detects suspicious activities such as fighting, weapon threats, fire, and accidents using real-time video feeds and alerts security teams. Built with TensorFlow, YOLOv5, and Raspberry Pi, it provides a cost-effective, automated solution for public safety.
🚀 Project Highlights
✨ Real-Time Suspicious Activity Detection
⚖️ Object & Gesture Classification with ANN & CNN
📹 Live Camera Feed or Pre-recorded Video Support
🏡 User-friendly Dashboard Interface
🎓 Educational Implementation with YOLOv5 + Google Teachable
🔗 Technologies Used
TensorFlow & TensorFlow Lite (model training and deployment)
YOLOv5 for object detection
Google Teachable Machine for ML without coding
OpenCV for video and camera integration
Matplotlib, NumPy, Scikit-learn, MPlayer
Raspberry Pi with Remote Desktop Interface
🔧 Features
🕵️ Suspicious Activity Detection
Detects: Fighting, fire, weapon threats, vandalism, and accidents
Uses camera feed or uploaded video for classification
🎥 Machine Learning Based
Artificial Neural Networks (ANN) and CNN used for training
Trained with labeled dataset using COCO + custom data
Pose estimation using PoseNet for behavior understanding
🚨 Alert System
Real-time alerts via sound and email
Alerts triggered if prediction confidence > 80%
📊 Admin Interface
Visual dashboard for monitoring alerts
Detailed activity info: time, location, type
Notification configurations
📈 Model Training Summary
Dataset: COCO + Custom labeled video frames
Tools: LabelImg, LabelMe, Google Teachable Machine
Training platform: YOLOv5 and Google Teachable
Epochs: 50+, optimized for underfit/overfit balance
📊 Results
Real-time detection accuracy: ~85-90%
Fast response with alerts under 2 seconds
Tested in environments with fire, fight simulations, object threats
🤞 Future Enhancements
Integration with drone and mobile surveillance
Use of multi-modal data (audio + video)
Auto-predictive analytics for threat probability
Real-time cloud sync and remote access
📚 Project Report Source
This README is based on the final year project report from DKTE’s Textile and Engineering Institute, Ichalkaranji.
📅 Developed With
Python
Raspberry Pi
TensorFlow
YOLOv5
Google Teachable Machine
OpenCV
📖 References
https://github.com/ultralytics/yolov5
https://teachablemachine.withgoogle.com
💼 License
MIT License
📢 Authors & Contributors Siddhant Gaikwad sandesh kamble Prasad Sabane
DKTE’s Final Year Engineering Team
Faculty Guide: [Faculty Name Here]
Contributors: