Accident Detection and Alert System

Road safety is a critical concern, with traffic accidents posing significant risks to human life and infrastructure. Accidents are considered one of the most serious problems for human population and cause great disruption to traffic flow worldwide. The effects especially loss of life and disruption of traffic could be addressed by timely detection of these events in real time and notifying relevant authorities. The goal of this proposal is to research on how to design and implement Automated Accident Detection and Notification System by using modern day image analysis and deep learning tools like CNN. Using cameras as input, the system detects unusual vehicle dynamics such as sudden stops and irregular vehicle movements that are precursors of an impending accident. In the case where additional learning is required to classify events, the proposed solution incorporates the use of OpenCV to preprocess and retrieve motion information from video sequences and a set of CNN models to classify the learned features of the encapsuled event video related to an accident. The system presented progresses the new studies in which, through the use of video, irregularities in traffic flows are quantified. Borrowing techniques of motion estimation like the Farneback Optical Flow and adaptive thresholding for anomaly detection, a solution to the problems is presented in the form of deep learning models. The objective is for accuracy though at low computation making it deployable in different settings like highways and urban intersections. The final goal of the project is improving road safety in general by fast-tracking responses during emergencies and limiting the aftermath of traffic accidents.