Text this: Deep Learning with a Spatiotemporal Descriptor of Appearance and Motion Estimation for Video Anomaly Detection