基于YOLO与改进的DLIB多角度遮挡人脸判别方法

Multi-angle Masked Face Discrimination Based on YOLO and Improved DLIB

针对ATM机上违法犯罪分子通过遮挡面部进行犯罪活动进而无法追踪的问题,提出了一种基于YOLO与改进的DLIB多角度遮挡人脸判别方法。通过将基于YOLO模型的多目标检测改成单一人脸检测,并调整其损失函数中人脸置信度损失计算方式,提高了人脸定位的准确性与时效性,完成了从原始图像的输入到任意人脸位置的回归,再结合改进的DLIB多角度人脸68个关键点检测算法在回归出的人脸位置上进行遮挡判别的新方法。测试结果验证了新方法优于传统方法,能够有效并快速地判别出各类遮挡,实现了ATM机上遮挡人脸判别的实时性与鲁棒性,具有较高的应用价值。

Aiming at the problem that the criminal criminals on the ATM machine can’t trace the criminal activity by covering the face,multi-angle masked face discrimination based on YOLO and improved DLIB is proposed.This paper proposes a new method of occlusion identification based on changing the multi-objective detection based on YOLO model into single face detection and adjusting the calculation method of face confidence loss in the loss function,the accuracy and timeliness of face positioning are improved,and completed from the original image input to any face position regression combined with68key point detection algorithms of modified multi-angle face in DLIB.The test results verify that the new method is superior to the traditional method,which can effectively and quickly identify all kinds of occlusions,and realizes the real-time and robustness of masked face discrimination on ATM machines.It has high application value.