Smoking Image Detection Based on Convolutional Neural Networks

• Published in: 2018 IEEE 4th International Conference on Computer and Communications (ICCC) pp. 1509 - 1515
• Main Authors: Zhang, Dongyan, Jiao, Cheng, Wang, Shuo
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.12.2018
• Subjects: CNN; Convolution; Deep learning; Feature extraction; image detection; Kernel; live streaming; streaming media; Testing; Training; Videos
• DOI: 10.1109/CompComm.2018.8781009

With the development of Internet technology and the improvement of network quality, online videos have become increasingly popular. In particular, online live broadcast has become a hotspot in recent years, and smoking behavior in these broadcasts is harmful to smokers and the surrounding environment. Therefore, it is necessary to detect and thereby effectively control smoking behaviors in video content. Traditionally, smoking images are detected based on the detection algorithms of cigarette smoke. Given the limited resolution of live broadcast videos, cigarette smoke is not visually apparent in the video content. This paper proposes a smoking image detection model based on a convolutional neural network, referred to as SmokingNet, which automatically detects smoking behaviors in video content through images. This method can detect smoking images by utilizing only the information of human smoking gestures and cigarette image characteristics without requiring the detection of cigarette smoke, showing high accuracy and superior performance for real-time monitoring.