Car accidents are perilous incidents occurring in all countries across the globe. Detection of such incidents plays a crucial role in improving road safety and facilitating timely emergency response. Hence, vehicle accident detection methods using road traffic videos are desired. In this study, we propose an effective car crash detection technique using a 3D Convolutional Neural Network (CNN) architecture. CNNs are able to extract crucial features from imagery and video data and use them to obtain excellent results in regards to classification problems. Our approach involves training a 3D-CNN to analyze spatio-temporal features in video data in order to detect car crashes in real-time. The data is taken from dashcams as well as surveillance cameras in order to cover multiple perspectives. The model achieves promising results in terms of multiple evaluation metrics representing a high accuracy and a low false alarm rate with the latter being difficult to decrease in these cases as can be seen by previous studies. Subsequently, the proposed 3D-CNN has the potential to enhance surveillance systems, aid in accident prevention and improve resource allocation for accident management.