Abstract:
Measurement system of exoskeleton robots can reflect the state of the patient. In this study, we combined an inertial measurement unit and a visual measurement unit to obtain a repeatable fusion measurement system to compensate for the deficiencies of the single data acquisition mode used by exoskeletons. Inertial measurement unit is comprised four distributed angle sensors. Triaxial acceleration and angular velocity information were transmitted to an upper computer by Bluetooth. The data sent to the control center were processed by a Kalman filter to eliminate any noise. Visual measurement unit uses camera to acquire real time images and related data information. The two data acquisition methods were fused and have its weight. Comparisons of the fusion results with individual measurement results demonstrated that the data fusion method could effectively improve the accuracy of system. It provides a set of accurate real-time measurements for patients in rehabilitation exoskeleton and data support for effective control of exoskeleton robot.
1. Introduction
Rehabilitation exoskeletons robot are emerging as important components of the rehabilitation training process for patients affected by hemiplegia and cerebral apoplexy [1,2]. The rehabilitation training system comprises a lower extremity exoskeleton, where the robotics are employed for sensing, control, information fusion and mobile computing [3,4,5,6]. Accurate data from measurement systems can be provided to the feedback control of exoskeleton system. By combining human intelligence with the physical exoskeleton, the robotic system can complete tasks via a man-machine interaction. Therefore, the methods used in measurement systems for obtaining lower limb gesture is of great significance.
Worldwide famous exoskeleton robots have been equipped with related measurement system. The measurement system in the ReWalk exoskeleton [7] developed by an Israeli researcher sends grit data obtained from a gyroscope sensor to a data processing center. The HAL exoskeleton developed at Tsukuba University in Japan, the measurement system is based on acquiring and analyzing EMG signals [8] as well as plantar pressure signals from the wearer, through dividing the gait it can control each phase while walking. Ekso [9], developed by an American company, uses crutches with attached sensors as well as plantar pressure shoes and an upper limb accelerometer to detect the walking intentions of the wearer. However, these detection systems focus only on a single mode during information acquisition and their accuracy is difficult to verify. In addition, the information obtained by these sensor systems exhibits hysteresis. Previous studies have shown that the attitude error increases as objects move, which can be eliminated by an external tracking system, such as a sonar, laser, or vision system [10] and an optimal motion model can be established by continuously updating motion information through a Kalman filter with a linear distribution [11]. Moreover, an unscented Kalman filter was proposed where the current state is considered based on a Gaussian distribution, thereby allowing multi-sensor data fusion [12,13].
But in different occasions, the measurement system plays different role. Many soft or flexible sensors change the measurement system. According to the different motion capture devices can be divided into mechanical motion capture, physical inertial sensor motion capture, acoustic motion capture, electromagnetic motion capture, optical motion capture and depth camera motion capture six categories [14,15]. Besides, types of sensors based on Micro-MEMS inertial sensing technology, such as Xsens, have been developed in order to obtain high-precision results, which can be applied to motion capture system [16]. Also, plenty of attempts for applying soft or flexible sensors to motion detection or monitoring system has been made, like a sensing system capable for monitoring human body [17] and soft sensors that can monitor the movement of the wearer and robot outside the lab [18].
The measurement system developed in the present study included an inertial measurement unit system for measuring human gait movement data [19,20,21] and a visual measurement unit system for acquiring real-time walking gestures from video image sequences. The inertial measurement unit system can obtain motion inertia parameters during walking, including from the hip joint and knee joint, thereby determining the movement posture and kinematics equation. The visual measurement unit system extracts and tracks feature point sets in the environment with a single camera and then calculates the position and pose of the robot with the measurement model and by extended Kalman filtering. The two methods for gesture data acquisition are supplementary and they can improve the reliability of the detection system. The final information obtained by data fusion [22,23] is sent to the robot information processing center of the lower extremity exoskeleton via wireless transmission to provide an experimental platform and theoretical foundation for intelligent walking and feedback control for the lower extremity exoskeleton robot, so the human motion can be measured in real time and the corresponding feedback control can be facilitated. The detection system is also based on our previous experiments where we aimed to improve the comfort and safety for users of our rehabilitation training exoskeleton [24].
In our experiment condition, we included the whole walking phase in our examination as mentioned. Because VMU system and IMU system can compensate for each other, the fusing results is applicable to all walking phases, including the even stance phase. Theoretically, the detecting method can be implemented without distance or velocity limit, however, thanks to the vision limit and potential dislocation of IMU, the perfect application condition is limited to a stride frequency range from 0.5 m/s to 1.0 m/s. According to existing research, such range is suitable for test subject, who are mostly slowly-walking stroke patients.
Figure 3. Marker points employs (R: right, L: left, H: hip joint, K: knee joint and A: ankle joint).
TBI Rehabilitation 