Improved MTCNN face detection method based on ShuffleNet

Abstract

The invention discloses an improved MTCNN face detection method based on ShuffleNet, and the method comprises the steps: transformation of different scales of an image is performed and an image pyramid is created so as to adapt to the detection of faces of different sizes. In the first stage, an original picture generates a face region Bounding Boxes through P-Net; in the second stage, the R-Net takes the original picture and the Bounding Boxes generated by the P-Net in the first stage as input, and generates the Bounding Boxes which are more accurate after correction; and in the third stage, the original picture and the Bounding Boxes output by the R-Net serve as the input of the O-Net, and the final face region Bounding Boxes are generated. A channel shuffling idea in ShuffleNet and a point-by-point layered convolution technology are adopted to improve the model. The model is based on the MTCNN, and the model is improved by adopting a channel shuffling thought during convolution operation, so that the network can quickly and accurately detect a human face.

Description

technical field [0001] The invention relates to the field of deep learning target detection, in particular to an improved MTCNN face detection method based on ShuffleNet. Background technique [0002] With the rapid growth of the number of motor vehicles, it has brought great convenience to people's life and travel, but the resulting road traffic accidents have caused huge losses to people's lives, properties and national economies in various countries every year. Fatigue driving is an important cause and one of the main causes of traffic accidents. If the efficient recognition of facial fatigue can be realized, the detection of the real-time facial expression status of the driver can effectively prevent and remind the driver of fatigue driving, thereby reducing the possibility of traffic accidents, so the system has potential economical value and broad application prospects. [0003] There are a series of problems in the existing driver fatigue detection methods: the dete...

AI Technical Summary

Problems solved by technology

[0003] There are a series of problems in the existing driver fatigue detection methods: the detection method based on physiological parameters requires the driver to wear intrusive experimental equipment, which not only affects the comfort, but also causes interference to the driver in the actual driving state

[0005] The proposal of AlexNet in 2012 opened the curtain of the development of deep learning, and then the proposal of VGGNet in 2014 made it possible to realize the deep neural network, but the problem of gradient disappearance will appear when the network is deepened.

Images