Coded Aperture Optimization for Single-Image Depth Thresholding Using Wearable Cameras

Cameras support many pervasive computing applications, such as activity recognition, but conventional imaging raises privacy concerns. Depth thresholding offers a privacy-preserving alternative by classifying whether objects lie within a specified distance range. However, accurate depth estimation typically requires intense computation or dedicated sensors, making wearable implementation challenging. This work proposes a method for depth thresholding using only a compact monocular camera by incorporating an optimized coded aperture (CA) into the lens. The method leverages the distance-dependent variation in the optical Point Spread Function (PSF) to determine whether a subject is closer or farther than a chosen depth threshold. This approach significantly reduces computational load and eliminates the need for specialized depth sensors, enabling low-power operation. Simulations with the optimized CA achieved a depth threshold classification accuracy of 68%, substantially outperforming conventional depth CA designs. These results demonstrate the promise of achieving accurate, privacy-preserving depth thresholding with compact cameras without requiring iterative computations.