Contrastive Research on Performance of Face Detection based on Classical and Deep Learning Algorithms

Abstract

Face detection technology has always been hot in the field of computer vision. However, current face detection algorithms are difficult to be applied to any scene with both high speed and accuracy. At present, there are few comparative studies between classical algorithm and deep learning algorithm. In this paper, our motivation is to study different types of face detection algorithm and make a comparison. We compare the average precision and receiver operating characteristic curve of RetinaFace with Mobilenet and Resnet. We also compare the performance of classical algorithm Viola-Jones with RetinaFace and deep neural network module in OpenCV library. We compare their detection performance in terms of computation time, frame rate, detection accuracy and model size under the conditions of multiple faces, high exposure and pose change. This study provides guidance and basis for researchers and developers to choose or design face detection algorithms.