Rapid Cigarette Detection Based on Faster R-CNN
Guijin Han, Qian Li, You Zhou, Jiawei Duan · 2019
Since the target detection algorithm based on deep learning is easy to be affected by light and image quality in cigarette detection applications, resulting the high false detection rate and high hardware occupancy rate, a rapid cigarette detection method based on Faster Regions with Convolutional Neural Networks (Faster R-CNN) model is proposed. This model based on the Faster R-CNN algorithm. Firstly, the algorithm of face detection is used to extract the range of face as the area of cigarette detection, therefore, the area of cigarette detection is reduced and many cigarette-like targets are filtered. At the same time, cigarette initial detection strategy based on color segmentation is introduced to judge the presence of a cigarette in face images preliminarily, after this step, the CPU occupancy and network complexity are reduced. The experimental results shows the proposed method reducing false detection rate of cigarette detection effectively and releasing the CPU occupancy rate greatly, and on the premise of ensuring the detection accuracy, the time of cigarette detection is reduced to some extent.