Occupancy Estimation from Blurred Video: A Multifaceted Approach with Privacy Consideration.

Area of Science:

• Computer Vision
• Building Energy Efficiency
• Human-Computer Interaction

Background:

• Accurate building occupancy data is crucial for resource allocation and emergency response.
• Traditional HVAC systems assume maximum occupancy, leading to significant energy waste (over 50% of US building energy budgets).
• Camera-based occupancy estimation offers high precision but raises privacy concerns.

Purpose of the Study:

• To develop and evaluate privacy-preserving occupancy estimation methods using intentionally blurred video frames.
• To investigate both motion-based and motion-independent techniques for occupancy counting.
• To analyze the trade-off between estimation accuracy and occupant visual privacy.

Main Methods:

• Proposed a privacy-preserving motion-based occupancy counting technique.
• Developed motion-independent methods including detection-based and density-estimation-based approaches.
• Utilized iterative statistical and deep-learning-based deblurring to enhance motion-independent method accuracy.
• Assessed privacy implications using image quality assessment metrics on original, blurred, and deblurred frames.

Main Results:

• The combination of iterative statistical deblurring and density estimation achieved a 16.29% counting error.
• This approach outperformed other proposed methods in accuracy.
• The study provided insights into the balance between occupancy estimation accuracy and visual privacy preservation.

Conclusions:

• A novel approach balancing occupancy estimation accuracy and occupant privacy was proposed.
• The iterative statistical deblurring with density estimation shows promise for privacy-aware occupancy counting.
• Further research is needed to fully optimize privacy-preserving occupancy estimation systems.