Road Accident Detection and Alert System at Real-Time Using Deep Learning Techniques

This paper presents a real-time road accident detection and alert system that leverages a custom sequential CNN model optimized for rapid accident identification. Unlike existing approaches, this method integrates an automated alert system through Telegram’s API, ensuring immediate notification to emergency responders. This enhancement significantly reduces response time, allowing for faster intervention and potentially saving lives. A convolutional neural network (CNN) model has been developed using frames extracted from CCTV video footage of accidents and non-accidents. The frames are processed and classified into an accident and non-accident categories for training the model to assess the incoming frames of a video in real-time. The data preprocessing, model selection, training, model validation, and performance assessment against benchmarks resulted in an efficient and accurate CNN-based accident detection system. It analyses live or recorded videos of the scene and determines whether an accident is present in the frame. In case of high confidence in an accident across multiple frames in succession, the system takes a picture in that frame, logs the time and location details, and then sends an alert message containing the time, location coordinates, and details of the accident to a specific Telegram channel through the Telegram Bot API. This monitoring method makes it easier to analyze the occurrence of an accident and report it within a short period.