Masked Face Detection Via a Novel Framework
Available Online March 2018.
- https://doi.org/10.2991/mecae-18.2018.137How to use a DOI?
- masked face detection; convolutional Neural Network; deep learning
- Masked face detection has a large variety of application like community policing, criminals capture, etc. Meanwhile, it is a challenging problem for academia. Study on normal face detection lasts for decades and has reached a high level in recent years. Contrastively, masked face detection still requires a lot of further study. Compared to the normal face detection, masked face detection is harder to deal with due to the loss of key points and diversity of occlusion degree. Traditional detection algorithm and framework work ineffectively on the problem. This paper researches into the problem, gathers related images, builds up a dataset, and proposes a novel framework for masked face detection. The whole system contains four modules. Proposal module produces candidate boxes and extracts the feature of each region with the help of two pre-trained network. Then, classification module trains a four-layer full connection neural network. The network is aimed at predicting five parameters of each proposed region. Regression module is designed to gain a more accurate position of the proposal. Finally, cluster module combines the information of neighboring boxes to give the final detection result. This module is different from the NMS algorithm which is frequently used in many traditional detection frameworks. Compared to the NMS algorithm, the new module produces a more accurate position and increases the robustness of the whole framework. Experimental results on the dataset show that the proposed framework remarkably outperforms 6 state-of-the-arts by at least 16.8%.
- Open Access
- This is an open access article distributed under the CC BY-NC license.
Cite this article
TY - CONF AU - Qiting Ye PY - 2018/03 DA - 2018/03 TI - Masked Face Detection Via a Novel Framework BT - Proceedings of the 2018 International Conference on Mechanical, Electronic, Control and Automation Engineering (MECAE 2018) PB - Atlantis Press SP - 238 EP - 243 SN - 2352-5401 UR - https://doi.org/10.2991/mecae-18.2018.137 DO - https://doi.org/10.2991/mecae-18.2018.137 ID - Ye2018/03 ER -