Anomaly Detection in Moving Crowds through Spatiotemporal Autoencoding and Additional Attention

Table 2

Comparisons with the state-of-the-art approaches in benchmarking datasets (frame-level AUC).

	Avenue	Ped1	Ped2
MPPCA [21]	N/A	59.0%	69.3%
MPPC + SFA [22]	N/A	66.8%	61.3%
Conv-AE [16]	80.0%	75.0%	85.0%
ConvLSTM-AE [18]	77.0%	75.5%	88.1%
Our approach with one ConvLSTM	82.9%	82.3%	89.3%
Our approach	85.7%	85.1%	92.6%