Runtime Monitoring of Accidents in Driving Recordings with Multi-type Logic in Empirical Models

Video capturing devices with limited storage capacity have become increasingly common in recent years. As a result, there is a growing demand for techniques that can effectively analyze and understand these videos. While existing approaches based on data-driven methods have shown promise, they are often constrained by the availability of training data. In this paper, we focus on dashboard camera videos and propose a novel technique for recognizing important events, detecting traffic accidents, and trimming accident video evidence based on anomaly detection results. By leveraging meaningful high-level time-series abstraction and logical reasoning methods with state-of-the-art data-driven techniques, we aim to pinpoint critical evidence of traffic accidents in driving videos captured under various traffic conditions with promising accuracy, continuity, and integrity. Our approach highlights the importance of utilizing a formal system of logic specifications to deduce the relational features extracted from a sequence of video frames and meets the practical limitations of real-time deployment.