In this paper, we propose a novel algorithm called sequence-based object detection to identify accidents in advance and prevent secondary accidents. To evaluate our proposed algorithm, we also propose a novel metric called sequence-based evaluation. We trained model for two datasets, one-class dataset and three-class dataset. For one-class dataset, we used focal loss, and as the results we achieved mAP@0.5 score above o.8 and accuracy 1.0. For three-class dataset, we compared focal loss and quality focal loss, and as the results we achieved mAP@0.5 score above 0.7 and accuracy 1.0.
