A crowd abnormal behavior detection method based on an autoencoder

Abstract

The present application belongs to the field of computer vision, and particularly relates to a crowd abnormal behavior detection method based on an auto-encoder. The present application extracts global context features by adding a global cross-channel feature extraction module in the encoder, extracts global space-time features by adding a global space-time information enhancement module between the encoder and the decoder, and simultaneously ignores irrelevant information and focuses on key information, so as to improve the detection capability of abnormal behaviors, improve the overall detection precision of the network, and be widely applicable to crowd abnormal behavior detection in videos.

Images