Tsagaantsooj Batzaya, Eun-bi Cho, Jeong-hyeon Chang, Kwanghoon Pio Kim, Journal of Internet Computing and Services, Vol. 26, No. 1, pp. 9-17, Feb. 2025
10.7472/jksii.2025.26.1.9, Full Text:
Keywords: CPTED, YOLO, GSV, Natural surveillance, Physical environment, Crime Prevention

Abstract

This study aims to build a deep learning-based YOLOv8 model that can detect streetlights, green spaces, and CCTVs as elements of natural surveillance in Crime Prevention Through Environmental Design (CPTED). Traditional measurement methods for natural surveillance have limitations in data collection. Therefore, this study progressively collected datasets related to streetlights, green spaces, and CCTVs provided by Roboflow, creating an integrated dataset composed of 13,715 images, which was applied to model training. The model was evaluated using 110 images captured at 50-meter intervals along specific streets in Ulaanbaatar, Mongolia, provided by Google Street View (GSV). Evaluation metrics such as Precision, Recall, and mAP were utilized. The results showed that Precision, Recall, and mAP@50 were above 0.50, while mAP@50-95 was below 0.50. The study concludes that the YOLOv8-based natural surveillance detection model effectively detects three environmental factors (streetlights, green spaces, and CCTVs) and is expected to be useful for crime prevention and urban hazardous environment data collection.

Statistics
Cite this article