An AI-Based Framework for Detection of Toxicity Levels from Oil and Gas Flaring from Video Streams

Industrial safety in the oil and gas sector is of great importance, particularly in relation to the environmental and public health risks associated with waste disposal through industrial chimneys. These emissions often contain hazardous pollutants contributing to air pollution, smog, and acid rain. The severity of these effects varies with toxicity levels, impacting both human health and ecosystems. To help industrial entities address these concerns and comply with international environmental regulations, this research presents a novel AI-based system suitable for real-time detection and analysis of flare and smoke characteristics. The solution overcomes the limitations of traditional sensors by providing the potential for integration with digital infrastructure for real-time alerts and reporting. The research provides the first annotated dataset for flare and smoke detection. It presents modeling for both day and night conditions using several prepossessing machine learning techniques combined with YOLOv5 and Fast RNN models. The models are tested on both images and video inputs, paving the way for effective deployment in industrial camera systems. The results demonstrate promising accuracy and performance outcomes for current and future directions.