Hand Gesture Recognition on a Resource-Limited Interactive Wristband

Area of Science:

• Human-Computer Interaction
• Wearable Technology
• Algorithm Design

Background:

• Existing hand gesture recognition algorithms demand high computational power, limiting their use in affordable consumer electronics.
• There is a need for efficient and accurate hand gesture recognition systems suitable for resource-constrained wearable devices.

Purpose of the Study:

• To propose a novel, computationally inexpensive hand gesture recognition algorithm for interactive wristbands.
• To evaluate the performance of the proposed algorithm against established methods like Dynamic Time Warping (DTW) and Recurrent Neural Networks (RNNs).

Main Methods:

• Fusing accelerometer and gyroscope data using a complementary filter to calculate three-axis linear acceleration.
• Defining a new feature, the 'axis-crossing code,' based on acceleration vector crossings in the world coordinate frame.
• Utilizing template matching for recognizing eight distinct hand gestures.

Main Results:

• The proposed algorithm demonstrated high accuracy (99.8% on collected data, 97.1% in user-independent tests).
• Achieved significantly lower computational resource requirements (Flash < 5 KB, RAM < 1 KB) compared to existing methods.
• Outperformed DTW in accuracy and significantly reduced time cost compared to DTW and RNNs (BiLSTM, GRU).

Conclusions:

• The developed hand gesture recognition algorithm is highly accurate, efficient, and suitable for consumer electronics.
• The algorithm's low resource requirements make it a competitive and practical solution for wearable devices.
• The technology has been successfully mass-produced and patented, indicating its market readiness.