Super-resolution human-silhouette imaging by joint optimization of coded illumination and reconstruction network

In surveillance camera systems and other human-image analysis systems, it is important to measure human shapes with high resolution. However, the performance achievable through approaches based solely on optical design and image processing has limitations. To overcome these limitations, we propose a super-resolution imaging system for human silhouettes based on a jointly-optimized design involving coded illumination and a reconstruction network. Our proposed method allows for the acquisition of human silhouette data with improved sampling resolution without modifying the camera hardware. We quantitatively demonstrated the effectiveness of our proposed method through simulations and also through optical experiments using a projector and a camera.