Table of Contents: :: Library Catalog

Saved in:

Bibliographic Details
Main Authors:	Dr. Naveen N C, Monisha H
Format:	Recurso digital
Language:
Published:	Zenodo 2026
Online Access:	https://doi.org/10.5281/zenodo.20210564
Tags:	Add Tag No Tags, Be the first to tag this record!

Table of Contents:

Abstract—The situational awareness under consideration requires the simultaneous situational awareness in a number of threat dimensions, both the behavioral conditions of individuals, and the physical position of dangerous objects. I am going to present in this paper one single unified and deep learning platform named SurveilAI which can be used to perform real time facial emotion recognition and weapon detection within a single web-deployable surveillance platform. The two perception tasks are reached by using single YOLOv8 models that are trained on domain-specific datasets. The trained weapon detection module, based on YOLOv8s and trained over 20 epochs on a Roboflow-curated dataset of four weapon categories i.e., grenade, gun, handgun and knife has a macro F1-score of 0.846 and an overall mAP50 of 0.922. With a per-class recall value of between 0.54 on contempt and 0.89 on sleepy, the emotion recognition module, which is based on YOLOv8n and trained on a nine-class dataset of facial expressions that include anger, contempt, disgust, fear, happiness, natural expression, sadness, sleepiness, and surprise achieves an overall F1-score of 0.750 with a maximally good confidence threshold of 0.371. Both models have a web interface implemented based on Flask with user authentication support, real-time image inference and FLASH notification of entities detected with bounding box overlays. Experimental testing testifies that the YOLOv8 architecture will be highly beneficial in terms of the simultaneous evaluation of behavioral and physical threats during surveillance systems, and the SurveilAI platform will offer a scalable and practical base of intelligent security applications

Similar Items