Face Detection Using Bionic Cascaded Framework

Face interaction plays an irreplaceable role in the service robots human-robot interaction, while face detection in this kind of scenario are challenging due to restrictions on computing capabilities and power, the character of real-time and requirements of the interaction pattern. Recent studies show that deep learning approaches can achieve impressive performance on these kinds of tasks. Therefore, a bionic cascaded framework adopted a cascaded structure with two stages of carefully designed face detectors is proposed in this paper, which exploits saccade and attention mechanism of human eyes to balance the performance. In addition, in the working process of the service robots, a new online sampling strategy that can improve the performance of interaction patterns is presented. In this way, the real time face detection and more natural human-robot interaction pattern can be achieved in service robot systems.