Integrated Multiscale Appearance Features and Motion Information Prediction Network for Anomaly Detection

Table 2

Frame-level AUC performance of different methods on UCSD dataset.

Methods	AUC (%)
	UCSD Ped1	UCSD Ped2
MDT [34]	81.8	82.9
Motion energy model [38]	75	81
Conv-AE [7]	81.0	90.0
ConvLSTM [8]	89.9	87.4
ConvLSTM-AE [9]	75.5	88.1
MGFC-AAE [10]	85	91.6
Unmasking [36]	68.4	82.2
AnomalyNet [37]	83.5	94.9
Baseline [15]	83.1	95.4
Proposed method	84.4	96.3