A New Method of Pedestrian Abnormal Behavior Detection Based on Attention Guidance

Table 3

Comparative experiments of the attention-directed abnormal behavior detection model (AUC%).

Method	Model	Ped1	Ped2	Avenue
—	Unmasking [28]	68.4	82.2	80.6
	Level set method [29]	79.3	89.8	—
	Two-stream R-ConvVAE [30]	75.0	91.0	79.6
	Two stream with OFF [24]	—	—	84.3
Reconstruction	CAE [31]	58.5	84.7	77.2
	AE_mem_recon [13]	66.3	89.3	80.5
	Att_AE_recon	67.6	90.7	82.7
Prediction	SRNN [32]	—	92.2	81.7
	GAN_pred [33]	82.4	93.0	84.6
	AE_mem_pred [13]	79.8	92.0	83.4
	ST-3DCAE [34]	80.7	85.3	81.0
	sRNN-AE [35]	—	92.2	83.5
	Two stream 3DAE [9]	79.6	90.3	82.0
	Memory_att [17]	—	—	85.7
	PST [36]	—	—	86.7
	Att_AE_pred	82.7	94.5	87.3