GssMILP for anomaly classification in surveillance videos

作者：

Highlights：

• Proposed a GSSMIL based video anomaly detection system.

• Presented a solution to resolve the HCI and LDoS problems in ACSV.

• Presented a C3D+3DCNN_BICLSTM architecture for effective feature extraction.

• Evaluated the proposed method on four datasets across 6 base-lines.

• Results indicate gain in f1-score between 9%–45% across 4 datasets.

摘要

•Proposed a GSSMIL based video anomaly detection system.•Presented a solution to resolve the HCI and LDoS problems in ACSV.•Presented a C3D+3DCNN_BICLSTM architecture for effective feature extraction.•Evaluated the proposed method on four datasets across 6 base-lines.•Results indicate gain in f1-score between 9%–45% across 4 datasets.

论文关键词：Anomaly detection,Multiple-instance learning,Graph-based semi-supervised learning,ConvLSTM,3DCNN

论文评审过程：Received 10 August 2020, Revised 19 April 2022, Accepted 27 April 2022, Available online 6 May 2022, Version of Record 13 May 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.117451