In India, accidents are a major cause of death. Over 80% of accident fatalities are caused by delayed assistance to victims. Accident victims can be left unattended for extended periods on lightly trafficked highways. To address this issue, we propose a system that uses deep learning to detect accidents from live CCTV video feeds. Each video frame is processed by a Convolutional Neural Network (CNN) trained to distinguish between accident and non-accident scenarios. CNNs are known for their speed, accuracy, and reduced preprocessing requirements, making them suitable for this task. With smaller datasets, CNN-based image classifiers have achieved over 95% accuracy.