CA-SSD-Based Real-time Smoking Target Detection Algorithm

Abstract

Smoking in public places is not only a key legislative project proposed by many countries, but also a hot topic in society. The traditional smoking ban supervision model has obvious disadvantages. For people s health and public safety. This paper uses an improved CA-SSD target detection model to detect smoking behavior, uses existing camera equipment in public places, and uses convolutional neural networks to extract, integrate and predict features. The images collected by the camera can accurately locate the smoker s position in real time. Aiming at the problem of too small cigarette butt targets and unobvious features, this paper improves on the original SSD network, uses resnet-50 as the feature extraction network, and integrates the feature fusion module based on CARAFE operator and the attention mechanism. Since there is no public data set, this article uses the smoking data set produced by ourselves, optimizes the parameters during training, and improves the accuracy of cigarette detection on the basis of ensuring real-time performance. Through the ablation experiment, the performance of the attention module and the feature fusion module is verified, which proves that the method in this paper has a better performance in the research of smoking detection.