Smart Retail Surveillance Reporter (SRSR)

Inspiration

Retail stores generate massive amounts of CCTV footage, but most of it remains unused. We were motivated to bridge the gap between passive surveillance and actionable insights, enabling retailers to understand customer behavior while preserving privacy.

What it does

Smart Retail Surveillance Reporter analyzes retail CCTV footage to:

• Detect customer–product interactions (picked, replaced, ignored)
• Track movement and engagement zones
• Generate automated behavioral insights for better layout and sales decisions
  

The system focuses on behavioral patterns, not personal identification.

How we built it

• Computer Vision: YOLOv8 and OpenCV for real-time detection and tracking
  
• Action Logic: CNN-based classification for interaction events
  
• Backend: Flask with a database to store and summarize events
  

At a high level, interaction likelihood is estimated as: [ P(\text{interaction}) = f(\text{hand proximity}, \text{object persistence}, \Delta t) ]

Challenges we ran into

• Handling occlusions and crowded scenes
  
• Differentiating between “viewing” and actual intent
  
• Maintaining real-time performance on limited hardware
  
• Ensuring privacy-aware analysis
  

Accomplishments that we're proud of

• Built a real-time, end-to-end retail analytics pipeline
  
• Converted raw video into structured, business-ready insights
  
• Designed a scalable and ethical surveillance solution
  

What we learned

• Real-world computer vision is significantly more complex than benchmarks
  
• Small logic tweaks can greatly improve accuracy
  
• Ethical considerations are critical in surveillance-based AI
  

What's next for Smart Retail Surveillance Reporter

• Live dashboards with heatmaps
  
• Predictive analytics for demand and shelf optimization
  
• Edge deployment for low-cost retail environments
  

Built With

• opencv
• python
• torch
• yolo