Guangyi Shi

Towards an ubiquitous motion capture system using inertial MEMS sensors and ZigBee network

In this paper, we developed a novel motion capture system with multi-sensing-node based on inertial MEMS sensors and ZigBee network, which is expected to be performed ubiquitously, not only strictly used in particular room as optical capture system does. First, the multi-node motion capture system was defined, which can track human motions and postures simultaneously and wirelessly. Second, multi micro inertial measurement units (uIMU) were designed by integrating the inertial MEMS sensors, micro controller unit and ZigBee module. At last, a wireless multi-sensing network was built up. Experiments showed that the sampling rate of this system was 50 Hz. The further work will be focused on the amelioration of ZigBee protocols for faster sampling rate of multi-node system, as well as human motion tracking algorithms.

Fall detection system based on inertial mems sensors: Analysis design and realization

This paper presents the development and analysis of inertial MEMS sensor based system that can detect falls in real time. The system is a major part of mobile human airbag system which prevents the elderly from fall induced fractures. The fall detection system hardware was designed, which could monitor the motions of the feet and waist and detect the falls in real time. Micro Inertial Measurement Units (μ IMUs) was applied in this system with Zigbee network and the fall detection algorithm what was constituted of three sub algorithms also was developed. The system was designed based on data analysis, in order to select the optimal parts for monitoring human motion and verify the algorithm performance, performance for different parts was compared by employing the pattern recognition based sub-algorithm and performance for different combination of human body segments and joints was also compared to get the better result. A wearable motion capture device was utilized to acquire the motion data. The effective extracting features were carried out and the motion classification performance was achieved and compared using the J48 decision tree classifier. Experimental results showed that the waist is the best location for motion monitoring with detection Sensitivity of 95.5%, the Specificity of 98.8% and the overall accuracy of 97.792%. Furthermore, the combination of the waist and feet sensing data was adopted with the Sensitivity of 98.9%, the Specificity of 98.5% and the overall accuracy of 98.565%. Based on the analysis, the system was designed to monitoring the motion of the combination, and the pattern recognition based sub-algorithm was also verified.

High accuracy Navigation System using GPS and INS system integration strategy

This paper proposes a position and attitude observer based on INS and GPS. Design and test results of an adaptive dual-rate Extended Kalman Filter(EKF) estimator for fusing data from Global Positioning Systems (GPS) and an Inertial Navigation System (INS) in order to estimate the position, velocity, and attitude. The dual-rate EKF consists of a high-speed filter and a low-speed filter, the high-speed filter fuses data from Real-Time Kinematic (RTK) GPS and INS, the low-speed filter fuses data from pseudorange GPS and INS. This solution designed to isolate the noise from pseudorange and realize the complementary of real-time performance and high precision. The solution yields exponential convergence of the attitude and position estimates. The implementation results show that the proposed method resolves an integer vector identical to that of the original method and achieves state estimation with centimeter global positioning accuracy.

Indoor positioning system based on inertial MEMS sensors: Design and realization

Nowadays, location-based services (LBS) has become widely used in our daily life. The most famous system is global positioning system (GPS), which is limited to outdoor applications and provide poor locating accuracy. In this paper, we present a positioning systems based on inertial MEMS sensor which includes three-axis accelerometer, three-axis gyroscope and three-axis magnetometer. The system can help people get accurate positioning for indoor environments, also available for outdoors, because of its self-contained character. It is a foot wearable device with wireless network to transmit movement information to computer that can calculate the relative position and show the path walked by. The key concept of the positioning system is inertial navigation and dead reckoning technology. Since it needs twice-integration of the acceleration to get the position, the displacement will drift by time elapse. We make it only drift by distance increasing through gait phase analysis, a method called Zero-Velocity Update (ZVU). As the “stand-still phase” is the key of the system performance, we mainly focus on getting accurate gait phase detection. We used decision tree here and the experimental results showed that we got a gait phase detection accuracy of 99.96% and positioning accuracy of 97.37%.

High precision tri-axial MEMS gyroscope module based on redundant implementation and sensor fusion

Aiming at the typical orthogonal configuration scheme (twelve sensors are orthogonal), the project implementation. With the redundant system and small guide system Navigation computer are linked together, complement a complete micro-miniature redundant strap-down inertial navigation system. Studied the system of fault detection, fault isolation and the system reconstruction technology, and use direct comparison measurement method and the weighted least squares method respectively as the fault detection and system reconstruction project implementation plan, implement the redundant in the system of IMU data collection, fault detection, identification, isolation and system reconfiguration, solving a series of functions such as navigation parameters.

Improvement of Chinese sign language translation system based on multi-node micro inertial measurement unit

This paper presents the improvement of Chinese sign language translation system based on studying in HMM algorithm, modeling and analysis. Based on the multi-node micro inertial measurement unit built by MEMS sensors, finger motion can be recoded and analyzed by computer. The computer preprocesses the data from the MEMS sensor nodes, including filtering noise removal and feature extraction. At last we train a classifier through HMM training process. Against 100 daily operations of sign language recognition experiments, the overall recognition rate is 90%. With the optimization and improvement of the algorithm, recognition accuracy and practicability will be greatly improved. We also designed a bidirectional translation system which can switch translation between Chinese sign language and voice freely.

Improvement of Chinese sign language translation system based on collaboration of arm and finger sensing nodes

With the rapid development of science and technology, the accelerated pace of life, people need to rely on language and hearing to exchange information, share information, learn and progress together. Currently there are about 70 million deaf and dumb people in the world, and in China there are about 20 million patients with different levels of hearing impairment. It is imminent to develop a set of devices that can connect the deaf and the normal people. With the development of human-computer interaction technology and pattern recognition technology, more and more researchers have entered into the field of sign language recognition and sign language translation. In this paper, we design a multi node sign language translation based on the pattern recognition technology. After data preprocessing, feature extraction, C4.5 decision tree algorithm recognition, 50 sign language actions with arms and body can get the recognition rate about 98%.

Improvement of human postural stability criterion using ZMP simplification and optimization algorithms

This paper presents a simplified Zero Moment Point algorithm for fall detection using inertial MEMS sensors. Based on body posture and stability judgment optimization system, we realized real-time detection and analysis of the falls. The system can be used in moving people airbag system that can prevent a major part of the induced fracture of the elderly from falls. Hardware includes the nodes that can monitor the movement of the feet and waist, and can detect normal motions and falls in real-time. The node is made mainly by Micro Inertial Measurement Unit (μIMUs). In this system, the detection algorithm is composed of three sub algorithms. The algorithms have also been developed and applied to Zigbee network and the fall detection. The system is based on data analysis and design, in order to choose the best means for monitoring the human movement, we verified the performance of the algorithm for different performance parts. Through the use of different combinations of segments of human pattern recognition algorithm, sub comparative performance was compared. Wearable motion capture device was used to acquire motion data. Based on this analysis, the system is designed to monitor a combination of movement, and the simplified zero moment point optimization algorithm has also been confirmed.

Algorithm of the fall prediction based on the double foot pressure and Micro Inertial sensors

This paper contains the development and analysis of the human motion state and the algorithm of the fall prediction based on the double foot pressure and the micro inertial MEMS sensors. The fall prediction hardware system consists of three parts, the double foot nodes and the waist node and how it was designed, which could measure the foot pressure parameters and the inertial parameters in different human motion states and could detect the falls in real time. Whats more, the foot pressure measurement units and the Micro Inertial Measurement Units (μIMUs) were applied in this system with wired network and the fall prediction algorithm was constituted of large numbers of threshold judgments which can detect different falls directly. With this hardware system, the foot pressure data and the motion data can be captured in real time. Then, these data will be dealt with through J48 decision tree classifier. Experiment results showed that the lead time (the time ahead of collision) of fall can be improved to 180ms and different falls can be recognized with different logic trees which were judged through the foot pressure threshold, the angular velocity threshold and the acceleration threshold. Based on the analysis, it can be showed that in the recognition of fall and ADL(Activities of Daily Life), the Sensitivity, the Specificity and the overall accuracy were all over 96%. While in the recognition of different falls, they can all achieve over 92%.

A fast magnetic deviation calibration algorithm for tri-axial electronic compass

According to the principle of electronic compass, we analyze the error of the E-compass and propose a calibration scheme in view of magnetic deviation, which is based on the self-designed tri-axial magnetometer system. The hardware is made by tri-axial magnetometer and tri-axial accelerometer. Considering the magnetic interference of the external environment will affect the magnetic deviation, we worked out a compensation method based on ellipsoid fitting. The experiment results showed the simplicity and efficiency of the algorithm. The compass deviation fell from 15 degree to within 1 degree. And it can maintain high accuracy even under large magnetic dip.