<![CDATA[Level crossings remain a problem in several cities due to high violations. Currently, surveillance at level crossings is still performed conventionally. Since problems at level crossings are increasingly complex and conventional solutions are no longer effective, an intelligent video surveillance system is necessary. Intelligent video surveillance system implementation is a complex task and requires devices with extensive computing resources. This research aims to optimize the system for processing data in real-time by conducting computation near the data source and dividing computing tasks across several edge devices. This research proposes a solution in the form of an edge computing-based intelligent video surveillance system with a computation offloading method on limited devices. This research has two development stages. The initial stage involved developing an object detection model using a dataset of level crossings in Bandung City. The second stage was developing an edge computing-based system by applying the computation offloading method on limited computing devices. The edge computing method extends cloud computing to the network’s edge, enabling calculations near the data source. Conversely, the computation offloading method improves edge computing performance by dividing computing tasks. Results showed an increase in computing speed of around 1.5 times faster, with a violation detection accuracy rate reaching 89.4%. Additionally, GPU temperature decreased by 5.50 °C, GPU usage decreased by 44.05%, memory usage decreased by 301 Mb, and power consumption decreased by 2.28 W. The system developed is effective and efficient in optimizing the performance of the violation detection system in level crossings on limited computing devices.]]>
Copyrights © 2024
