Spatio-temporal Analysis of Dashboard Camera Videos for Time-To-Accident Forecasting

Globally, traffic accidents are the leading cause of death according to reports published by the World Health Organization (WHO). Advanced Driver Assistance Systems (ADAS) are effective in reducing the likelihood or severity of accidents. However, such systems are expensive and inaccessible to the majority of the population. Time-To-Accident (TTA) forecasting is a critical component of ADAS systems as it can allow for better decision-making in traffic, improve dynamic path planning, alert the driver of potential dangers, and more. Sensor or depth imaging-based TTA forecasting can be inaccurate and are unaffordable to the mass. Existing vision-based traffic safety research in the literature focuses on accident anticipation with limited work on TTA prediction. We propose forecasting TTA by utilizing spatio-temporal features from dashboard camera videos and introducing a novel algorithm to compute ground truth TTA values. Additionally, we present a CNN-Transformer Hybrid (CTH) model that overcomes the limitations of end-to-end CNN or Transformer-based models. Our best models can predict TTA values with an average prediction error of 0.24 and 0.80 seconds on the CCD and DoTA dataset respectively. Furthermore, We assess our approach's ongoing scene recognition ability using the TTA-ASiST@x (Time-to-Accident-based Accuracy at the
$\mathrm{x}^{th}$
frame Since Scene Transition) metric. In real-world scenarios, it is vital to identify when a normal driving scene transitions into a potential accident scene. With just 25% of the data, HyCT achieved over 85% accuracy in recognizing accident and non-accident scenes, further showcasing its performance on limited data.