| Summary: | Traditional security surveillance detection relies on post-event forensics or is hosted on a backend server, making it impossible to identify behaviors filmed in the field online. This paper proposes the Smart City Security System (SCSS) for detecting anomalous activity in public locations online. SCSS combines the DeepSORT and YOLOv4 algorithms to generate the DS-YOLO aberrant behavior detection algorithm, which compares and matches the target detected in the previous picture frame with the target detected in the following frame to achieve detection and tracking. SCSS is equipped with GPS, WIFI, and Uninterruptible Power Supply (UPS). When a risky behavior is detected, the system will upload the abnormal event as well as the latitude and longitude that occurred to the cloud via the WIFI and notify the user. The recognition accuracy of three deviant behaviors, including Fight, Car Accident, and Fall, was examined using diverse situations, and the results were 89%, 90%, and 90.33% respectively. The findings demonstrate that SCSS has successfully made the transition from passive monitoring to active identification, offsetting the flaws of conventional security systems that can only post-mordem forensics, and bridging the gap of the construction of national smart cities.
|
|---|