Case study of face detection libraries on Raspberry Pi

With the fast development in recent years of micro PC, applications that require of using computer vision techniques have become more and more popular. This paper describes a thoroughly research on four different face detectors using the two most popular computer vision libraries (i.e., OpenCV and Dlib), and compares the detecting speed of those detectors. The face detectors, which are examined in all experiments, include two classic detectors (Haar and HOG based face detector) and two newer, deep neural networks based detectors. Also, four different Raspberry Pi models including Raspberry Pi Zero W, Pi 2, Pi 3, and Pi 4 are put into the experiments of face detecting speed. The results show that the Raspberry Pi 2, Pi 3, and Pi 4 are capable of doing real-time face detection tasks using Haar or HOG based face detector, and all four Raspberry Pi models can be used in practical applications where face detection is required. For other detectors based on deep neural networks, although are not suitable for face related real-time tasks, they are still effectively used in many other face related computer vision tasks. Keywords: Face Detection, Raspberry Pi, OpenCV, Dlib.