Research on Face Detection Algorithm Based on SSD Algorithm
Aiming at the problem of poor small target detection effect in the traditional SSD algorithm,a face detection algorithm based on ResNet is proposed.This paper improves the basic network VGG of the SSD algorithm to the ResNet network,and uses the method of characteristic integration to integrate the characteristic information of different depth through the residual network,thereby improving the detection performance of the small-scale face.At the same time,aiming at the problem of leak detection in the overlapping frame for the SSD algorithm,the Non-Maximum Suppression(NMS)is improved to Soft-NMS.In addition,by setting up an attenuation function,it reduces the confidence of the adjacent detection box,and solves the problem of filtering the traditional NMS algorithm on the detection box with lower scores,which could reduce the leakage rate of the algorithm and improve the detection accuracy of the algorithm.
face detectionSingle Shot MultiBox Detector algorithmResNetSoft-NMS
万方数据
维普
