P-FAD: Real-time face detection scheme on embedded smart cameras

Face detection is known to be a resource-hungry process in computer vision. Most of existing detection algorithms, such as Viola-Jones detector, are too time-consuming to be implemented on resource-limited embedded smart cameras. Observing that the computation increases proportionally to its pixel manipulation, and inspired from the cascade structure which imposes more complex processing on the promising regions, hierarchical approach may be a feasible solution. The challenge issue in the design is how to construct a multi-layer architecture, in which the complex processing can be split from the pixel manipulation and guarantee detection accuracy simultaneously. To remedy this problem, we propose a novel Pyramid-like FAce Detection (P-FAD) scheme that results in a significant reduction of computation during face detection process. P-FAD has five-layer architecture, in which the operating units decrease dramatically from top to down while imposing complex computations in the last layer. We have implemented our scheme both on a notebook and our embedded smart camera platform. Our experimental results demonstrate that the P-FAD schemes resource-aware properties while still hold the acceptable detection accuracy.