<![CDATA[Currently, street crime remains a serious challenge in Indonesia, while conventional CCTV systems still function passively as recorders. One of the most concerning types of crime is robbery with violence, commonly known in Indonesia as begal, which remains among the most frequently reported cases. This study proposes the Begal-Detector, a YOLOv8-based system that integrates Human Activity Recognition (HAR) and Object Detection to identify suspicious activities in real time on edge devices. The experiments were conducted on Raspberry Pi 4, Raspberry Pi 5, and Raspberry Pi 5 with Hailo AI Kit, with variations in distance, camera angle, and lighting conditions. The test dataset consisted of 72 video samples, including both street crime and non-street crime scenarios, recorded using the EZVIZ H8C Outdoor CCTV camera. Experimental results show that the Begal-Detector performs very well, achieving a 100% detection accuracy at a distance of 2 meters, 94% at 3 meters, and 94% at a 45° camera angle. Under low-light conditions supported by infrared light, the system maintained an accuracy of up to 79%, making it feasible for real-world deployment. In terms of hardware performance, the Raspberry Pi 5 with Hailo AI Kit provided the most optimal results, achieving an average of 52.71 FPS with a stable temperature of 63 °C, significantly outperforming the Raspberry Pi 4 and Raspberry Pi 5 without an accelerator, both of which failed to operate the system in real time. The findings confirm that utilizing Raspberry Pi 5 with Hailo AI Kit is an effective solution to ensure that the Begal-Detector operates quickly, stably, and reliably.]]>
Copyrights © 2025
