Abnormal Event Detection in Surveillance Videos Using Two-Stream Decoder

Abnormal event detection in surveillance videos refers to the identification of events that deviate from the normal pattern. An autoencoder can be used to learn the normal patterns from the videos, and its reconstruction errors can be used to detect the abnormalities. Surveillance videos consist of two components: dynamic objects and a static background. Because of the nature of the static background, we can assume that the source of abnormality is from the objects. In this work, we propose the use of a two-stream decoder model to tackle the abnormal event detection problem in surveillance videos. The two-stream decoder comprised a background stream that models the static background and a foreground stream that models the dynamic objects. We also utilized a two-stream encoder to learn from optical flow, which contains motion information, and skip connections used to improve the details in the output frames. Several experiments on publicly available datasets were used to validate the effectiveness of the proposed model.