In contemporary times, many companies are investing in autonomous vehicles, with several models already on the market and their usage increasing daily. Smart vehicles utilize various sensors to enhance passenger safety, but challenges remain, particularly for individuals with dyspnea or motion sickness. Cigarette smoke, which contains harmful substances like carbon monoxide (CO), poses risks to passengers and can interfere with the vehicle’s sensor ecosystem due to its odor and particulate matter. Detecting cigarette smoke is challenging due to its varying textures, colors, and shapes. To address this, we propose a Deep Learning (DL)-based detection system using Transfer Learning (TL). A pre-trained VGG19 network, originally trained on ImageNet, was fine-tuned for feature extraction and trained on a cigarette smoke dataset. Our model achieved a classification accuracy of 96.05%, surpassing those of existing architectures. Additionally, we compared our model to current architectures and satisfied our expectations.