Authors:
Pavithra Babu, A. Shree Harini, S. Sowmiya
Addresses:
Department of Computer Science and Engineering, Sathyabama Institute of Science and Technology, Chennai, Tamil Nadu, India.
Abstract:
Hazard-to-first-responder time affects the security and survival of international individuals. Traditional emergency response systems work, but victims need independence, a voice amplifier, or a way to call for help. Domestic violence, human trafficking, and fast jail threaten victims. Lightweight, wide computer vision platform CovertSign AI wants smart CCTV cameras for emergency response. The global Signal to Help, a hand motion to avoid aggressors, is recognised in real time. The MediaPipe Hands 3D landmark extraction model converts visual hand movements into 63-dimensional feature vectors for 21 major joints. For reliability and cheap processing cost, Researchers use a machine learning classifier ensemble. Automated Hand Gesture Recognition and Performance Analysis: Our gradient-optimised Random Forest model outperformed SVM and K-Nearest Neighbour classification methods on over 1,000 samples across varying lighting and distances. GMM and SVM optimise 30 fps edge computing on consumer-grade hardware with 23% CPU consumption. Automatic multi-channel warnings are issued when a distress signal is at least 90% certain. Local visual detection and automated high-priority SMTP email warnings with live timestamps and high-resolution scene captures are used. From gesture start to emergency notice, the pipeline takes 5–10 seconds. With its hardware-agnostic, cheap solution, CovertSign AI can help vulnerable people in residential, business, and public spaces switch to passive monitoring and positive intervention. Improved gesture recognition pipelines enhance technological accessibility and detection precision to protect humans.
Keywords: Hand Gesture Recognition; Random Forest; Precise Detection; Technological Accessibility; Positive Intervention; Passive Monitoring; Live Timestamps; Edge Computing.
Received on: 06/12/2024, Revised on: 25/01/2025, Accepted on: 20/04/2025, Published on: 03/01/2026
DOI: 10.64091/ATICL.2026.000253
AVE Trends in Intelligent Computer Letters, 2026 Vol. 2 No. 1 , Pages: 26-41