Smoker Detection Image Dataset

Dataset Curation

This dataset was meticulously curated by utilizing various search engines with multiple keywords such as “cigarette smoking,” “smoker,” “person,” “coughing,” “taking inhaler,” “person on the phone,” and “drinking water.” The intent was to gather a versatile range of images that create a degree of inter-class confusion, thereby enhancing the robustness of machine learning models trained on this data.

Class Descriptions

• Smoking Class: Contains images of individuals smoking from multiple angles and performing various gestures associated with smoking.
• Not Smoking Class: Includes images of individuals engaging in activities with gestures similar to smoking, such as drinking water, using an inhaler, holding a mobile phone, and coughing. This intentional overlap aims to challenge and improve the model’s discriminative capabilities.

Image Specifications

• All images are preprocessed and resized to a resolution of 250×250 pixels.
• The dataset is divided with 80% of the images allocated for training and validation, and 20% for testing.

Usage and Applications

The Smoker Detection Image Dataset is a valuable resource for researchers and developers aiming to develop and refine deep learning algorithms for:

• Automated Detection: Enhancing public health by identifying smokers in public spaces.
• Environmental Surveillance: Supporting green initiatives through monitoring and reducing smoking-related pollution.
• Smart City Implementations: Integrating with smart city infrastructure for improved surveillance and enforcement of no-smoking zones.