Face Detection of Innovation Base Based on Faster RCNN

In order to improve the efficiency of teachers’ attendance and save teachers’ energy and time, this paper proposes a face detection algorithm based on improved Faster-RCNN, in order to overcome the problems of gradient disappearance and gradient explosion caused by too deep network depth. In this paper, ResNet50 is used to replace VGG16 backbone feature extraction network, and soft non-maximum suppression method is used to improve the recognition rate of overlapping faces based on the principle of Faster-RCNN algorithm, ResNet50 residual network model and RPN network structure principle are described in detail. Training and testing on Wider Face data set, carrying out a variety of network comparison experiments, provide a strong basis for the experimental results. At last, the experimental data show that ResNet50 feature extraction network is relatively ideal and the detection accuracy on the test set reaches 85.3%, which is 3.9% higher than that of VGG16 on average. The detection time of a single picture is 0.34 s, which meets the real-time monitoring requirements and provides a new idea for the face attendance system.