Cover Story：Dai T, Zhang K, Yin D. An end-to-end neural network for mobile phone detection in driving scenarios[J]. Opto-Electron Eng, 2021, 48(4): 200325

Nowadays, mobile phones have become an important part of people's life. A lot of people are inseparable with mobile phones, and the phenomenon of using mobile phones while driving is common, which is dangerous. Therefore, real-time monitoring of driving behavior is not only beneficial to the control of the traffic management department, but also of great practical significance to reduce the occurrence of traffic accidents. Nowadays, with the rapid development of computer vision, especially the rise of convolutional neural network, image and video processing is more accurate, efficient, and more widely used.
An end-to-end neural network for mobile phone detection in driving scenarios was proposed by Dong Yin teams, associate professor of Intelligent Information Processing Laboratory, School of Information Science and Technology, University of Science and Technology of China. In the method, an end-to-end small target detection network OMPDNET is designed to extract image features, which ensures the real-time processing while maintaining a high recognition accuracy. In order to generate the anchor suitable for small target data, improve the efficiency of the network model and overcome the problem of noise data and hard example data, they improved the K-means clustering algorithm and proposed a clustering algorithm K-means-Precise, which is more suitable for the distribution of sample data. In addition, they used supervision and weak supervision to construct their own data set, and added negative samples to the data set for training. Experimental results show that the proposed method has good recognition accuracy and fast processing speed, which is better than many mainstream target detection algorithms at present. At the same time, this work have a certain expansion and inspiration for deep learning in small target detection.