Efficient Hand Gesture Recognition Using Artificial Intelligence and IMU-Based Wearable Device

Gesture recognition is an important element of human–computer interaction that allows natural and intuitive communication in applications such as healthcare, rehabilitation, smart home environments, safety, gaming, and accessibility solutions for individuals with disabilities. The electromyography (EMG) and mechanomyography (MMG) sensor-based traditional approaches suffer from limitations such as noise susceptibility, critical placement requirements, and inefficient detection of broader arm movements. Further, they do not work for individuals with amputation or minimal muscle movement, as muscle activity is not available. Addressing these challenges, herein, we present a novel wearable hand gesture recognition system which is less prone to noise and placement issues. The presented devices use accelerometers and gyroscopes to capture hand and arm gestures. Further, the developed wearable system employs 1-D convolutional neural networks (1-D CNNs), long short-term memory, and recurrent neural networks for efficient processing of data and recognition of gestures. The 1-D CNN with three convolutional and three dense layers emerged as the optimal solution, achieving an accuracy of 97.88% with balanced inference time and memory usage. The study concludes that this model offers an optimal trade-off between model size and accuracy, making it highly suitable for resource-constrained wearable devices.