Concealed objects detection for human security surveillance system
Loading...
Date
item.page.authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The threat of terrorism remains a major concern across the globe.
newlinePublic transport, social events, iconic locations, crowded places, critical
newlineinfrastructure are vulnerable targets. In addition to loss of human life, it raises
newlineand instills a fear about public safety. In most cases the tools or devices are
newlineconcealed and escapes the surveillance systems. So it is essential to detect
newlineconcealed objects under human clothing using an advanced security
newlinesurveillance system. Very few studies have been conducted on hidden or
newlineconcealed object detection. Most of these studies faced several struggles in
newlineobtaining better accuracy and performance. To address these issues, a
newlineModified Weighted You Only Look Once v5 (MWYOLOv5m) model is
newlineproposed in Phase 1 and YOLOv5m model in Phase 2.
newlineIn Phase1, weighted boxes fusion (WBF) method is employed to
newlinediminish the inaccurate results generated due to low confidence scores. The
newlineCSPDarknet53 acts as a feature extractor and the Path Aggregation Network
newline(PANet) implements feature aggregation. The YOLO hyperparameter is tuned
newlineusing a crossover salp swarm algorithm (CSSA) to provide better accuracy in
newlinefinding hazardous weapons. The datasets utilized for the MWYOLOv5m
newlinemodel are Millimeter wave (MMW) and Terahertz (THz) datasets. The
newlineimplemented model is compared with the state of art models such as Faster
newlineRCNN, SSD, and YOLOv5m. The precision acquired using MMW and THz
newlinedatasets is 98.97% and 97.15% respectively.
newline