A Deep Learning Based Automatic Outdoor Home Video Surveillance Approach
摘要
Home video surveillance systems are widely used to track potential criminal activities and serve as a formidable deterrent to potential intruders and wrongdoers. Nonetheless, it’s essential to emphasize that these systems very frequently lead to false alarms, which can be quite frustrating for users. The existing automatic video surveillance methods and datasets were initially developed for monitoring public crimes, and they do not generalize well for home surveillance purposes. To this end we proposed a deep learning based automatic home video surveillance approach based on anomaly detection. We captured a representative vide dataset for our outdoor home surveillance system and employed this dataset to train our deep learning model. Our evaluations have shown that our model achieved anomaly detection accuracy of 79.12%, which is impressive considering our dataset size.