Real-Time Smoking Detection System

Computer vision and deep learning pipeline for real-time smoking behavior detection from camera input.

Problem

Detect smoking behavior in live video under varying lighting conditions, using face/hand cues rather than only coarse scene classification.

Approach

• Face and hand landmark detection with OpenCV and CVZone
• CNN classifier trained/served with TensorFlow for smoking behavior recognition
• Geometric cue: distance between hand and face landmarks to improve decision quality
• Hardware enhancement: infrared / thermal sensing path to improve reliability under difficult lighting and heat conditions

Tech stack

• Python
• OpenCV
• CVZone
• TensorFlow (CNN)
• MATLAB (analysis / supporting experiments)
• Camera + IR / thermal sensing modifications

Results

• Real-time detection pipeline from live camera stream
• Landmark-aware smoking behavior classification
• Improved robustness through IR/thermal sensing support

Detection examples

Smoking detected	No smoking	Heat detected

Hardware / setup

Demo

Watch demo video

My role

End-to-end CV/ML pipeline design, model integration, landmark-based logic, and sensor/camera hardware modifications for reliability.

Relevance

Demonstrates applied ML on visual edge-style pipelines using TensorFlow and OpenCV — directly relevant to Edge AI perception work.