Energy-Accuracy Aware Finger Gesture Recognition for Wearable IoT Devices

Area of Science:

• Computer Science
• Electrical Engineering
• Human-Computer Interaction

Background:

• Wearable Internet of Things (IoT) devices are crucial for gesture recognition.
• Balancing high accuracy and low energy consumption in these applications is challenging.

Purpose of the Study:

• To design an energy-accuracy aware finger gesture recognition system using wearable IoT devices.
• To optimize system-level performance and energy consumption for low-end microcontrollers.

Main Methods:

• Developed a finger gesture recognition system employing a lightweight multi-layer perceptron (MLP) classifier and a 2-axis flex sensor.
• Created system-level performance and energy models for accuracy and energy optimization.
• Analyzed design choices and identified Pareto-optimal solutions for embedded wearable IoT devices.

Main Results:

• Achieved up to 95.5% gesture recognition accuracy.
• Demonstrated energy consumption below 2.74 mJ per gesture on low-end embedded wearable IoT devices.
• Provided 159 design choices for energy-accuracy aware design points.

Conclusions:

• The proposed system effectively achieves high accuracy and low energy consumption for finger gesture recognition on wearable IoT devices.
• The energy-accuracy aware design framework is suitable for low-end microcontrollers.
• Pareto-optimal designs offer practical solutions for resource-constrained embedded systems.