Learning Deep Representations of Appearance and Motion for Anomalous Event Detection