Experimental Study on Multiple Face Detection with Depth and Skin Color

Abstract

Human face considers as an important biometric trait for person identification or video surveillance due to the digital camera technology that available on our daily life gadget. Since the digital signage easily found in the public and uncontrolled environment, the common situation could be single or multiple audiences that viewing at the digital signage display. The digital camera acts as a non-invasive detector for non-obtrusive digital advertising to collect the surrounding people’s face. The accuracy rate of face detection becomes the priority to detect the audience’s face. Besides that, the processing time also become a concern for time-constrained applications. This paper develops a framework for non-obtrusive digital advertising that applied the depth camera to detect multiple audiences for the audience location simulation and gather the depth information restrict the region of interests (ROI). Viola-Jones algorithm detects the audience frontal face who is facing towards the digital signage in the ROI. Subsequently, skin color analysis verifies the skin face and exclude the non-skin face to improve the face detection true detection rate. The depth information is combined with the face XY-position to map the audience actual location in the real-world environment on the aerial map. The experiment result shown that post-processing approach for Viola-Jones algorithm with skin color analysis increases the face true detection rate with the short processing time. Meanwhile, the simulation of multiple audience locations in the ROI can be shown on the aerial map which corresponds to the actual location in the real-world environment.