This study proposes the ROS 1 framework and a 2D LiDAR employed two algorithms based on particle filtering for SLAM (Simultaneous Localization and Mapping). The algorithms include gmapping (Grid-based FastSLAM Mapping) and adaptive Monte Carlo localization (AMCL) to establish maps and determine the positions for the robot navigation. Gmapping disperses particles generated by the 2D LiDAR throughout the environment, calculates particle weights, and then constructs a map by drawing these particles. AMCL matches them with known maps to estimate the robot's position. The navigation is carried out using the move_base for the analysis of six outdoor positioning points of the 2D LiDAR navigation robot, and the comparison uses RTK-GPS as the reference point. The target aims to measure the errors of the 2D LiDAR navigation robot concerning these six points. The comparison shows that the root mean square (RMS) errors of RTK-GPS are consistently smaller than those of 2D LiDAR.
Incidences of stroke among young individuals have been increasing, highlighting the importance of rehabilitation therapy for future recovery. Rehabilitation scoring scales have been widely used in the field of rehabilitation to assess various abilities. However, these scales often fail to provide a comprehensive evaluation of each patient's specific condition, as they may miss certain specific movements or situations. To address this issue, this study aims to establish a Stroke Rehabilitation Exercise Assistance System based on Openpose. The system effectively identifies crucial movements encompassing sitting posture, the process of transitioning from sitting to standing, and standing posture, achieving F1 Scores of 0.966, 0.947, and 0.706, respectively. The research involved 17 stroke patients with an average age of 68.5 years. Over a three-month period, a total of 77 test sessions were conducted. By employing the Pearson correlation coefficient, the study analyzed the experimental results from these sessions. The findings revealed a significant correlation (84%) between the time taken by stroke patients to perform the sit-to-stand movement and their scores on the Berg Balance Scale, indicating a strong association between improvements in rehabilitation movement time and the individual's balance function and fall risk. The proposed model's accuracy and correlation with balance function measurements show promise for enhancing rehabilitation outcomes and patient care in the field of stroke rehabilitation.
Air pollution has the characteristics of multiple sources, various scales, and cross-regional distribution, making it impossible to be solved solely through pollution prevention and treatment in individual counties or cities. Therefore, for the study of air pollution, Sequence pattern mining is suitable to analyze the relationship between air pollution and regions. Due to the fact that most previous research has mainly used the level intervals of individual air pollutants and their monitoring values as the basis for regional correlations, rarely utilizing the level intervals of multiple air pollutants along with the monitoring values of a single air pollutant as the basis for regional relationships, this study uses the data of PM 2.5 , PM 10 , and 8-hour moving average of ozone (O 3 -8Hr) with Air Quality Index (AQI) sub-indices greater than 101 in the southern region of Taiwan from 2020 to 2022 as the basis, aiming to uncover relationships in the southern region of Taiwan and contribute to future air quality management.
This paper aims to study the different control methods and implementation of the dual-drive gantry control system. In this paper, the Euler-Lagrange method is used to derive the mathematical dynamic model of the system. The derived dynamic model is then linearized and decoupled for the control design. Simulink is used to conduct simulation studies of the designed control systems. In the mechanism design of the dual-drive gantry system, the system can be easily converted into a soft connection or a rigid connection enabling us to study the effects between these two structures. The control laws are implemented in the experiments through a digital signal processor. The decoupling control and parallel control methods are used to control the dual-drive gantry system with a load placed in different positions. The performance comparison of different control methods is made. Finally, it is shown that the proposed decoupling control method can reduce the load disturbance. Moreover, the proposed decoupling control method has a lower tracking error when the system is in the soft connection.
In recent years, convolutional neural networks (CNNs) have been widely applied in emotion recognition. To improve the large parameter configuration and time-consuming problems, a smart embedded emotion recognition system is proposed and implemented in this study. Firstly, a modified OpenPose is proposed to lighten the computational load. One stage of Part Affinity Fields (PAFs) is reduced, resulting in three stages for PAFs and two stages for key joints. After extracting the required keypoints, 21 vector magnitudes are calculated as the emotion features real time. Then a backpropagation neural network (BPNN) is designed as a classifier. To train the model, all the 21 vector magnitudes are used as inputs for the BPNN. The recognized results of the classifier contain six emotion categories. Finally, the proposed system is implemented in the NVIDIA Jetson Xavier NX as a smart embedded system. Some experimental results are provided to prove its effectiveness. The future goal is to apply the embedded recognition system into companion robot to interact with the host.
