Research on Face Detection Algorithm Based on Yolov3
Aiming at the problems of low precision,low recall and slow speed on face detection in the complex real scenes,this paper proposes a face detection algorithm GL-Yolov3 based on improved Yolov3.The algorithm adopts the depthwise separable convolutions and the standard convolutions construct the feature extraction network(Darknet-MI),the GIoU as loss for bounding box regression is utilized to design the loss function of the algorithm.At the same time,the K-means++ clustering algorithm is used to analyze and set the size of the anchor box to better adapt various face detection scenarios.In addition,the feature fusion algorithm based on efficient channel attention mechanism is proposed,which can obtain more useful feature information by recalibrating the different channels of feature map.The average precision of the proposed algorithm on the Wider Face dataset is 91.7%,and the aver-age detection time is only 24.3ms.It has been proved by experiments that the improved Yolov3 algorithm effectively realizes re-al-time face detection under the premise of ensuring high accuracy.
face detectionK-means++depthwise separable convolutionsGIoUECA-Net
国家科技期刊平台
NSTL
万方数据
