Women’s safety in hostels, campus walkways, and other remote public areas is based on the ability of a surveillance system to identify a developing issue. This paper we present a study of YOLOv8 and Faster R-CNN as the vision components for Sentinel, our real-time multi-modal women’s safety system which includes features like weapon detection, face covering detection, acoustic distress analysis, and motion sensing which we have put together for CPU-only laptop hardware. We also report our implementation and evaluation of YOLOv8 in the knife and face covering modules of Sentinel, also we present as a control the use of a pre-trained Faster R-CNN ResNet-50 which we used as a drop-in alternative and tested on the same knife and scissor set we used for the primary evaluation. In our in-house test we found the YOLOv8 to have a 95% detection rate at a speed of 31.4 ms per frame (31.8 fps), in that which Faster R-CNN reported 75.0 percent accuracy at 194.2 ms per frame (5.1 frames per second) we see 20 percentage point greater accuracy and approximate 6.2x lower per-frame latency which in turn supports YOLOv8 as the better base for Sentinel’s CPU-only real-time human-in-the-loop safety pipeline.

yolov8,Faster-RCNN,women safety