The Internet-of-Things based hand gestures using wearable sensors for human machine interaction

Hand gestures for human machine interaction using wearable sensors have more potentiality than ambient sensing thanks to its low-cost, light weight and mostly scalablity every where at anytime. Despite the fact of existing works on human hand gestures using wearable sensors, each focuses on a specific application and difficult to be generalized. In addition, it still lacks an available benchmark of hand gestures in the context of human machine interaction. This paper introduces a new human hand gesture dataset which could be suitable for controlling home appliances. The dataset is captured with a low-cost and sensor plugable Internet of Things (IoT) device which is currently embedded with accelerometer and gyroscope sensors. We then investigate various features extracted from multiple sensor data for training several machine learning models. Furthermore, we propose a simple yet effective late fusion model from multimodal data for enhancing the recognition rate. In our preliminary experiments on the collected dataset, we demonstrated that the proposed late fusion schema considerably improves the accuracy of gesture classification. The highest accuracy achieved with late fusion technique is 87.61%. These results are highly promising for practical applications that utilize human gestures such as controlling of the appliances at homes and human machine interaction.