Development of a Wearable Sign-language Recognition Prototype with Dual LCD Feedback Using Flex Sensors and IMU Technology

Abstract

Aims: This engineering study aimed to develop and evaluate a low-cost wearable sign-language recognition prototype for controlled recognition of selected isolated American Sign Language (ASL) alphabet and number gestures using finger-flexion and hand-orientation signals. The system was not intended to provide complete natural-language interpretation. Study Design: Prototype development and controlled bench and gesture testing of a sensor-integrated glove system. Place and Duration of Study: STEM Science Center/student engineering laboratory, New Jersey, USA; project period: 06/2025 to 03/2026. Methodology: Five 2.2-inch resistive flex sensors were stitched into a dual-layer glove and mechanically stabilised with hot glue and strain-relief routing. Each channel was connected to the analogue input of an Arduino Mega 2560 through voltage-divider/amplifier circuitry. An Arduino Nano 33 BLE Sense Rev2 mounted on the dorsum of the glove provided inertial measurement unit (IMU) data for orientation-sensitive gestures. Sensor readings were calibrated using rest and bent positions, smoothed, baseline-corrected, and interpreted using threshold- and rule-based mapping to gesture labels. The translated output was displayed on mirrored 20 x 4 I2C LCD screens, with serial/SD logging used during calibration and troubleshooting. Results: Four of the five flex channels produced stable, repeatable, and separable motion signatures suitable for controlled isolated-gesture testing. Baseline levels for F1-F4 were high and distinct (approximately 630-845 ADC counts), with low baseline noise for F2-F4 (0.50-0.80 ADC counts). The F5 channel showed extreme baseline instability (SD 57.79 ADC counts; CV of approximately 132.9%), consistent with intermittent wiring, floating input, or divider instability. IMU roll and pitch signals provided orientation context that complemented flex-sensor patterns. Formal classification accuracy, confusion-matrix analysis, and multi-user generalisation were not claimed from the present dataset. Conclusion: The prototype demonstrates that a low-cost microcontroller-based glove can capture interpretable multimodal signals for selected isolated ASL alphabet/number feedback under controlled conditions. The strongest engineering contribution is the integration of flex and IMU signals with dual LCD feedback and the systematic identification of channel-level reliability limits that must be addressed before larger-scale validation.