Accelerating Facial Detection for Improvement of Person Identification Accuracy in Entering and Exiting Management System

摘要

Recently, needs of individual personal entering and exiting information are high by development of deep learning. In order to automatically acquire entering and exiting, it is necessary to acquire a facial image of a human who enters or exits within about one second, so a faster face detection technology is required. And the installation space of photography equipment is narrow. We developed a high-speed facial area estimation algorithm by reducing the facial search area using a high-speed image processing and speeding up with GPGPU. By executing on GPU of Jetson TX 2, execution time of the facial area estimation becomes about 14 ms and accelerating rate with respect to the conventional method is 60 times. This result shows that practical facial area estimation processing is possible even on an inexpensive and compact processor.