Overview / Usage

Overall Project Description: The project leverages advanced computer vision techniques to identify smoking instances in images and videos. It utilizes two key technologies: YOLO (You Only Look Once) and Haarcascade. YOLO is a deep learning algorithm capable of detecting objects in real-time, while Haarcascade is a feature-based object detection method. These technologies work synergistically to ensure accurate and efficient smoking detection.

Problems Being Solved: Smoking is a widespread health concern that poses severe risks to individuals' well-being. By detecting smoking instances, the project addresses several problems:

1. Health Hazard: Smoking is known to cause a wide range of health problems, including lung cancer, heart disease, and respiratory disorders. Detecting smoking can encourage individuals to reconsider their behavior and make healthier choices.
2. Public Awareness: Providing real-time feedback about smoking instances raises public awareness about the hazards of smoking. This can influence not only smokers but also those around them.
3. Behavioral Change: Visual cues can be powerful triggers for behavioral change. Detecting smoking instances in real-time can serve as a reminder and incentive for individuals to quit or avoid smoking.

Experienced or Used in Production: The project has practical applications and can be implemented in various scenarios:

1. Public Spaces: The technology can be integrated with security cameras in public spaces such as malls, airports, and educational institutions. When smoking is detected, security personnel or management can be alerted to take appropriate actions.
2. Health Campaigns: Health organizations can use the technology in their awareness campaigns. Videos and images can be analyzed to demonstrate the dangers of smoking, driving home the message with real-time detection.
3. Smartphone Apps: The project's capabilities can be integrated into smartphone apps. Users can receive alerts and notifications when smoking is detected nearby, helping them avoid exposure.
4. Research: The project can be a valuable tool for researchers studying smoking behavior. Data collected from real-world scenarios can provide insights into smoking patterns and its impact on society.

Methodology / Approach

Methodology Overview:

1. Data Collection: A diverse dataset of images and videos containing smoking instances and non-smoking scenes is collected. This dataset serves as the foundation for training and testing the detection model.
2. Preprocessing: The collected data undergoes preprocessing, including resizing, normalization, and augmentation. This step ensures that the data is suitable for training and improves the model's generalization capability.
3. Model Training: YOLO (You Only Look Once) and Haarcascade are employed as the primary detection models. YOLO is a state-of-the-art deep learning algorithm known for its real-time object detection capabilities. Haarcascade, on the other hand, is a feature-based object detection method. Both models are trained on the prepared dataset to recognize smoking instances.
4. Integration with Flask: Flask, a micro web framework for Python, is used to create a user-friendly web interface. The integration allows users to upload images and videos for smoking detection. Flask also enables real-time notifications when smoking is detected.
5. Real-time Detection: YOLO and Haarcascade models are integrated into the Flask application. The uploaded images and videos are processed using these models to identify smoking instances.
6. Alerts and Visualization: When smoking instances are detected, the application provides real-time alerts to users. Additionally, the application can overlay visual indicators on the detected smoking areas, enhancing awareness.

Technologies and Techniques Used:

1. YOLO (You Only Look Once): YOLO is a deep learning model that performs object detection in real-time. It can identify multiple objects in a single pass and is well-suited for applications requiring speed and accuracy.
2. Haarcascade: Haarcascade is a feature-based object detection technique. It uses a set of trained features to detect objects, making it efficient and reliable.
3. Flask: Flask is used to create a web-based interface for user interaction. It facilitates the integration of the detection models and allows real-time alerts and visualization.
4. Data Augmentation: Data augmentation techniques, such as image rotation, flipping, and color adjustments, enhance the diversity of the dataset, leading to better model performance.
5. Pretrained Models: Pretrained YOLO and Haarcascade models are used as a starting point. Fine-tuning these models on the smoking dataset accelerates the training process.
6. Web UI Design: User interface design principles are applied to create an intuitive and user-friendly web interface for uploading and analyzing images and videos.

Technologies Used

Technologies and Libraries:

1. Python: The primary programming language used for coding the project.
2. Flask: A micro web framework used to create the user interface and handle HTTP requests.
3. OpenCV: An open-source computer vision library used for image and video processing.
4. YOLO (You Only Look Once): A deep learning framework for real-time object detection.
5. Haarcascade: A feature-based object detection method included in OpenCV.
6. NumPy: A library for numerical computations in Python.
7. PIL (Pillow): Python Imaging Library for image processing tasks.
8. Matplotlib: A data visualization library used for creating plots and graphs.

Tools and Software:

1. Anaconda: A Python distribution that includes various data science libraries.
2. Jupyter Notebook: An interactive coding environment for data analysis and experimentation.
3. Visual Studio Code: A code editor with built-in Git integration and extensions for Python.
4. GitHub: A version control system for collaborative development.
5. Intel DevCloud: An Intel platform for testing and optimizing deep learning models.

Hardware:

1. CPU: A powerful processor capable of handling image and video processing tasks.
2. GPU (Graphics Processing Unit): Accelerates deep learning model training and inference.

Other links

• github (https://github.com/sufyn/Smoking_detection)

• webapp