A New Video-Based Crash Detection Method: Balancing Speed and Accuracy Using a Feature Fusion Deep Learning Framework

Table 2

Crash detection models’ performances on testing set.

No.	Model	True positive (TP)	False negative (FN)	False positive (FP)	True negative (TN)	Accuracy (%)	FPS
1	Faster R-CNN + SORT + rules [24]	58	32	18	72	72.22	0.73
2	ResNet-50 + CBAM + rules	69	21	12	78	81.67	50
3	ResNet-50 + CBAM + LSTM	70	20	12	78	82.22	27
4	ResNet-50 + SE + Conv-LSTM	74	16	12	78	84.44	35
5	ResNet-50 + CBAM + Conv-LSTM	78	12	11	79	87.22	33
6	ResNet-50 + CBAM + Bi-Conv-LSTM	79	11	11	79	87.78	30