No abstract

No abstract

Human-movement recognition is a novel challenge in soft robotics. In recent years, there have been several attempts to develop soft wearable devices for supporting human-robot interfaces. Many algorithms and programming languages are available to integrate a wearable device with a soft robot. One such promising algorithm is reservoir computing (RC), which includes of a group of recurrently and randomly connected nodes. The RC model can easily process multidimensional signal data and can handle nonlineardata and has been extensively used in robotic control. It has been reported that the RC algorithm can speed up network training and solve complex data sets. However, the main existing limitations in handlocomotion classification are the considerable run-time and the delayed response. In this study, we figure out the best machine learning algorithms to handle three-dimensional hand-gesture data. We employ a two-part strategy: a loopback filter is included in the preprocessing of the initial dataset to support the 3dimensional (3D) signs of each hand posture; subsequently, the training dataset is applied to the machine learning algorithm which includes an artificial neural network (ANN), convolutional neural network (CNN), long short-term memory(LSTM), and reservoir computing(RC). Each training network is optimized with various hyperparameters. Furthermore, we compare the performance of several machine-learning algorithms in classifying the three-dimensional hand-signal posture data. The results show that the classification of nonlinear hand-locomotion signals by RC requires a comparatively shorter training duration (12 minutes for training times), and that optimal accuracy 94.17, precision 94.10, and recall 93.99 is realized for time series data.INDEX TERMS Human-machine interface, human hand-locomotion signal, reservoir computing, time series, multi dimension, and nonlinear data.

In recent years, the digitization of documents has progressed, and opportunities for handwritten document creation have decreased. However, handwritten notes are still taken for memorizing data, and automated digitalization is needed in some cases, such as making Excel sheets. When digitizing handwritten notes, manual input is required. Therefore, the automatic recognition and input of characters using a character recognition system is useful. However, if the characters are inclined, the recognition rate will be low. Therefore, we focus on the inclination correction problem of characters. The conventional method corrects the inclination and estimates the character line inclination. However, these methods do not work when characters exist in independent positions. Therefore, in this study, we propose a new method for estimating and correcting the tilt of independent handwritten digits by analyzing a circumscribed rectangle and other digital features. The proposed method is not based on an AI-based learning model or a complicated mathematical model. It is developed following a comparatively simple mathematical calculation that can be implemented on a microcontroller. Based on the results of the experiments using digits written in independent positions, the proposed method can correct the inclination with high accuracy. Furthermore, the proposed algorithm is low-computational cost and can be implemented in real-time on a microcontroller.

Visually impaired individuals worldwide are at a risk of accidents while walking. In particular, falling from a raised place, such as stairs, can lead to serious injury. Therefore, we attempted to determine the best accident prevention method that can notify visually impaired individuals of the existence, height, and step information when they approach stairs. In this study, we have investigated stair detection through deep learning. First, the three-dimensional point cloud data generated from depth information are learned by deep learning. Stairs were detected using the results of deep learning. To apply the point cloud data for deep learning-based training, we proposed preprocessing stages to reduce the weight of the point cloud data. The accuracy of stair detection was 97.3%, which is the best performance compared to other conventional methods. Therefore, we confirmed the effectiveness of the proposed method. INDEX TERMSVisually impaired support systems, Depth sensor, 3D point cloud data, deep-learning, PointNet I. INTRODUCTION