Emergency Gesture Recognition System Using Deep Learning: A Case Study Based on an ASL Hand-Sign Detection Pipeline

Introduce an Emergency Gesture Recognition System (EGRS) for critical commands when speech fails; quantify training-to-deployment performance gap.

Existing ASL models lack real-world reliability. This project offers compact, domain-specific emergency gestures with realistic evaluation.

Signaling for help in disasters, industrial/medical emergencies; enabling surveillance systems, rescue robots, drones, and wearables.

. In critical emergency situations such as natural disasters, industrial accidents, or medical emergencies, normal speech-based com- munication might not always be reliable. Background noise, damaged devices, or physical injuries can prevent people from speaking or being heard. In such cases, hand gestures offer a reliable alternative to sig- nal for help. We introduce an Emergency Gesture Recognition System (EGRS), a lightweight, high-accuracy gesture recognizer built specifi- callyforemergencyscenarios. Thesystemadaptsaconvolutionalneural network (CNN) originally designed for American Sign Language (ASL) recognitionandrefinesittoasmaller, targetedsetofemergency-related gestures. After addressing dataset duplication artifacts discovered dur- ingvalidation, ouremergency-onlymodelachieves 97.42%testaccuracy on a cleaned, held-out emergency test set (736 images). In a separate realistic live-webcam evaluation, the system achieved 51.4% accuracy, demonstrating a substantial training-to-deployment gap that we ana- lyze in detail. A critical contribution of this work is the identification and quantification of this performance gap between laboratory evalua- tionandreal-worlddeployment, emphasizingtheimportanceofrealistic evaluation protocols in gesture recognition research.