TMCnet News
An Indoor Positioning Method with MEMS Inertial Sensors [Sensors & Transducers (Canada)](Sensors & Transducers (Canada) Via Acquire Media NewsEdge) Abstract: A new method is put forward to realize pedestrian's localization by inertial sensor in indoor environment, the positioning method is through pedestrian's motion information which is collected by an IMU system to achieve pedestrian's positioning. The pedestrian's gait is getting by fusion of acceleration data and gyro data, it is improved the traditional method which discriminate pedestrian's gait by single axis acceleration. A new method is proposed for estimating the pedestrian's walking step length by acceleration data, it is improved the estimation of the dynamic step length precision. By linear fitting algorithm to correct the yaw angle information, it would calculate more accurate yaw angle information. Using the step number, step length and yaw angle by fusion algorithm reproduce a pedestrian's route. The experimental results show that the positioning method of current positioning precision for each 1200 m error is in 10 m. Positioning error is about 8.33 %, it has high value of engineering application. Copyright © 2013 IFSA. Keywords: Indoor positioning, Inertial measurement unit system, Trajectory reconstruction, Positioning precision. (ProQuest: ... denotes formulae omitted.) 1. Introduction In recent years, data services and multimedia services are rapid increased, demand of positioning and navigation are increasing, especially in complex indoor environment, such as airport hall, exhibition hall, warehouse, supermarket, library, underground parking, mine environment and so on, it is often need to determine the mobile terminal or the holder of the indoor position information. The GPS positioning technology is relatively mature, but in the indoor conditions, due to the received GPS signal is very weak or no receives, the GPS will be out of action. According to statistics there are 80 % to 90 % of people's time is indoors, so indoor positioning technology become a very hot topic. In recent years, including Google, Microsoft, Apple, Broadcom and some of the world famous university are in the study of indoor positioning technology, many indoor positioning technology solutions were proposed, but by the positioning time, positioning accuracy and complex indoor environment conditions, perfect positioning technology solutions is still not well used in the actual application. Due to the MEMS inertial sensor has the advantages of miniaturization, multi-function, low cost [1], it is widely used in the inertial navigation technology and it becomes the focus of research of the indoor positioning technology [2], The collecting device used in this paper has a very high level of integration, it is not only has the advantages of small volume and convenient carrying, and high precision data acquisition, at the same time the sampling frequency also can be set by friendly interface, through the pedestrian's walking posture information which collected by the collecting device, we can make data processing to achieve walking trajectory and then realize the reappearance of pedestrian's walking trajectory for pedestrian localization. We present a new kind of indoor positioning method of multiple MEMS inertial sensor based on data fusion. The article is organized as follows: the second section describes the IMU system and the collection method, the third section presents the gait identification method of the pedestrian, the fourth section is step length estimation method and solving of pedestrian's yaw angle, the fifth section gives the trajectory reconstruction method and test result, the sixth section is the conclusion. It is the intelligent combination of multiple MEMS sensors which consist of 3 axis accelerometer, 3 axis gyroscope, 3 axis magnetometer, thermometer and pressure sensors. The data processing chip is 32 bit STM32 microcontroller, the sensing function of the system is acceleration, angular velocity, magnetic, temperature and pressure, it is 11 degrees of freedom. At the same time, the corresponding input parameters in the IMU interface of the system also can obtain the yaw angle, roll angle, pitch angle and the four element data, it provides a convenient for us to achieve positioning. The fixation method of IMU system has great effects on the measured data, some researchers bind the IMU system on their toes [3]. The method is feasible, but noise of the measured data is a little big, so that data processing is more difficult. In this design, the IMU system is fixed on the tester's right leg, let the tester's direction coincide with the Z axis of the IMU, the X axis direction is perpendicular to the ground as far as possible, the Y axis direction is parallel to the ground and perpendicular to the surface of the testing process, as shown in Fig. 2. In this test, the pedestrian advancing direction is Z axis negative direction, the Y axis of the gyroscope is pedestrian's horizontal axis, that it is perpendicular to the direction of the plane which is consist of the direction of forward and vertical to the ground plane. 3. Identification of Gait The walk distance of pedestrian are usually integral the acceleration of every step, so it need to get the accurate number of steps of the pedestrian, there are many method to get the number of steps [4-7], to judgment the tester whether static or dynamic is by accelerated speed such as Koichi [4], he believes that the tester is dynamic if the value exceed the range he has set; Ojeda [5] judge it by the threshold of the angle speed; Cho [6] judge it by the rate of change of the acceleration, if the rate of change is greater than a certain threshold, that means the sensor is in motion. These methods are belongs to threshold method, it set the variable threshold in advance to get the number of steps, but there are many factors influence variables, such as the rate of change of acceleration, angular velocity and so on, in a mass of test, we found that just use a single threshold method often occur omission or error, such as Fig. 3, the abscissa represents the sampling points, the blue line is the direction of the acceleration, usually use it to judge the pedestrian whether static or dynamic, the threshold method can achieve the right number of steps if it is able to determine the A point, B point, C point and some of these special points. But it is clearly that the threshold method is unable to achieve the ideal effect, C point will be miss detection, D point and E point is very easy to false detected, it will lead to errors of getting number of steps so that the location accuracy will be descend. In order to make the process method is more reasonable, we propose a new method that added to the gyro data as an auxiliary judgment, gyro axis Y data can be a very good response of walking state. When the Y axis is perpendicular to the surface of pedestrians, you can see the leg of pedestrian is a regular swing from the Y axis direction. The data obtained from gyro of Y axis shows at MATLAB software is the red line in Fig. 3. Experiments show that the data has a high reliability and good stability. In Fig. 3, there are two kinds of different wave of the red line, such as F and G, in order to find the maximum value of F, we can set a reasonable threshold, the threshold value is greater than G value and is less than F value, we set it to 1/2 of F value and G value that can reduce the sentence range and reduce the amount of calculation. Then find the value which is greater than or equal to the previous and less than or equal to the behind. Then it is the point of maximum point at F. Gyro value is the value of the instantaneous angular speed, the maximum angular rate value at F is the biggest moment of every step, it is not the right leg landing time but the time before. Because the difference of the time and the landing time is relatively stable, we can use this time as a reference time, then find the maximum acceleration point after a period of time. If find the maximum acceleration point, the time of the point is the right leg landing time, if not, replace it with the experience value. To find out the right leg landing time, we can accurately calculate the total number of steps, the program flow chart is shown in Fig. 4. 4. Step Length Estimation and Yaw Angle Calibration 4.1. Step Length Estimation There are many estimation methods of step length, Koichi [4] uses a three axis accelerometer and a single axis gyroscope fixed on the shoes to estimate walking distance, in the total distance of 30 m, the maximum error is 5.3 %. Ojeda [4] uses six degrees of freedom inertial measurement unit estimates the walking distance, walking distance error within the 600 m is less than 2 %. But the drift error of gyroscope will accumulate and then it will affect travel distance results in the process of testing, so the error will increase with the increase of travel distance. Pointed out in document [5] that walking distance in 600-620m the maximum error is 12.8 % and in order to reduce the test error needed the tester every 30 s to stop a time, so that calibrate the gyroscope. The gyro error accumulate easily [8], the peripheral effect of metal material and the price is not for most people to accept, so people consider using acceleration sensor with the advantages of small size and low price to estimate the distance. The parameters of getting the number of steps are fixed so that the measurement results relate to the speed and the lean of road. It need neural net to estimate the distance of every step, but the parameters need a large number of test data to determine and there are some difference between learning data and actual data lead to the system error increases [9]. Aiming at the problem, a new step length estimation algorithm was proposed. The walking process can be decomposed into three parts, the leg lifting, walking and hitting the ground. The walking process can be simplified, we can approximate it as a uniformly accelerated rectilinear motion, the leg stop quickly when it hits the ground and the time is very short that can be ignored. The step length of this pedestrian can be calculated by the formula of displacement and acceleration. As shown in formula (1), where 'S' represents step length, 'a' represents acceleration. ... (1) H - - - In Fig. 3, A point and B point can be seen as the leg lifting time and hitting the ground time respectively. The front half part can be regarded as static of the right leg. The latter part can be regarded as right leg produced displacement. Set a threshold aO of the acceleration, when the data above it means pedestrian moving, otherwise stationary. The section of the acceleration value is larger than ao were ai, a2, a3,..., an, (a0, ah a2, ..., an only represents the acceleration value) and then calculate the average value of am of the acceleration, its mean value formula shown as formula (2), ...(2) We can see that is decided by walking speed of the pedestrian. By formula (1) we can see that step length can be calculated by acceleration and time. Suppose acceleration is ar, time is t. Because the average acceleration value am can be only get in the actual situation, and can't get the value of ar, we need a method to get the value of ar. Through the analysis we can find that ar and am have a common characteristic that if pedestrian is stationary, the two values are approximate to 0 and if walking faster, the two value is greater, if moving slower the two value is smaller, assumes the ar and am is a linear relation, it can be written as shown in formula (3): ... (3) where A is the proportional coefficient, it can be obtained through the measurement. In the experiment by limiting step, repeated testing, we calculated that the A value is about 0.5 and remain relatively stable. Through A value and the sampling points, sampling frequency in each step, we can get the value of ar, and the walking time t in this step, substitute ar and t in formula (1), then we can calculate step length of this step. The tester walked on the football field about 15 minutes, the total number of steps is about 800. The step length is about 1.5 meter after the measurement. Substitute the measured data in designed algorithm, we calculate the step length as shown in Fig. 5, where abscissa represents the walking number of steps, ordinate represents step length, sign represents the direction, the test result shows that the calculated step length is with high accuracy and good stability. 4.2. Yaw Angle Calibration The yaw angle is one of the important elements in positioning system, the definition is that the right, front, top direction of the carrier constitute the righthanded coordinate system, around the forward axis rotating is roll angle, around the axis towards the right rotating is pitch angle, around the upward axis rotating is yaw angle [10]. So for pedestrians, yaw angle is the walking direction information. The front of the pedestrian is 0°, around the clockwise rotating angle value is positive, the range is 0°-180°, so the yaw angle range is -180°-180°. Through the friendly interface of IMU system, we can get the yaw angle information after we input the correct parameters, due to the influence of vertical gyro drift and axial gyro drift [11] and so on, it will cause the major angle error. After the tests we also found that there are big errors between the measured yaw information and actual yaw angle, so we need to calibrate attitude angle, fixed the test device on the two-axis turntable and collected the attitude angle data. The model of two-axis turntable we used in the test is 902-E which is made by Beijing Aviation Precision Machinery Research Institute. The test results are in Table 1. As can be seen from the Table 1, there is a certain bias value between the reference value and system output, it is accord with linear rule, therefore make the system output value and reference value for the linear fitting and then we can get more accurate yaw angle value. 5. Trajectory Reconstruction Through the above discussion we have introduced how to calculate the three factors [12] - number of steps, step length and yaw angle, the principle of trajectory reconstruction is that regard the pedestrians walking in plane trajectory as composed of many points, as shown in Fig. 6. Where aK means the K step yaw angle, Sx means the X axis coordinate of pedestrians at that time, SY means the Y axis coordinate of pedestrian at that time. You can write a coordinate of pedestrian on the coordinates of any point of the plane, the calculation formula (4) is as follows. ...(14) According to the theory above, this paper conducted a number of experiments, the experimental one is in a hall and the walking tracks is layout good as shown in Fig. 7. The left two pictures are field test photos and the two on the right picture are trajectory which gets from the test data, the top one is walking track 1 lap, following is continuous walking track 5 laps, the blue 'x' represents the point of fall right, as can be seen from the chart, trajectory and actual trajectory reconstruction basic consistent, continuous test results can still accurate response to walking trajectory with good repeatability and stability. In order to further test the positioning accuracy, the test site of experiment two is a standard football field, the tester along the soccer field periphery of the 400 m runway walks and collects test data. Fig. 8 is a map of walking trajectory reconstruction after three laps, the abscissa and ordinate units are meters, pedestrian walked a distance of about 1200 m, the positioning accuracy is within 10 m, and have good stability. 6. Conclusion In this paper, a new indoor data fusion positioning method based on multiple inertial sensors was proposed, it has made overall and deep analysis of the three elements of dead-reckoning method involved, especially on getting the number of steps and step length, we make a bold improvement and innovation. The experimental result shows that, within walking distance of about 1200 m, the positioning accuracy is within 10 m, the positioning error is 8.33 %, greatly improves the positioning accuracy of the pedestrian. Acknowledgements This work was financially Supported by the National Natural Science Foundation of China (No. 51175335). The National Science and Technology Innovation Fund Projects (No. 10C26215113044). The Science and Technology Project Affiliated to the Education Department of Chongqing Municipality (No. KJ110507) and Chongqing Natural Science Fund Project (No. CSTC-2012jjB40003). References [1] . D. H. Titterton and J. L. Weston, Strap down inertial navigation technology, 2nd Edition, American Institute of Aeronautics and Astronautics, 2004. [2] . D. H. Titterton and J. L. Weston, Strap down inertial navigation technology, 2nd Edition, American Institute of Aeronautics and Astronautics, 2004. [3] . J. Bird and D. Arden, Indoor navigation with footmounted strap down inertial navigation and magnetic sensors, IEEE Wireless Communications, Vol. 18, No. 2,2011, pp. 28-35. [4] . S. Koichi, I. Hikaru, S. Yutaka, Non-restricted measurement of walking distance, in Proceedings of the IEEE International Conference on System, Man and Cybernetics, Nashville, 2000, pp. 1847-1852. [5] . L. Ojeda, B. Johann, Non-GPS navigation with the personal dead-reckoning system, in Proceedings of the SPIE, Orlando: Unmanned Systems Technology, 2007, pp. 6561-6572. [6] . Seong Yun Cho, Chan Gook Park, Measurement system of walking distance using low-cost accelerometers, in Proceedings of the 4th Asian Control Conference, Singapore: ACPA, 2002, pp. 16-21. [7] . A. T. M. Willemsen, F. Bloemhof, H. B. K. Boom, Automatic stance-swing phase detection from accelerometer data for personal nerve stimulation, IEEE Transactions on Biomedical Engineering, Vol. 37,No. 12, 1990, pp. 1201-1208. [8] . Senog Yun Cho, Chan Gook Park, MEMS based pedestrian navigation system, The Journal of Navigation, Vol. 59, No. 1, 2006, pp. 135-153. [9] . Q. Yang, L. Cheng, G. C. Chen, Estimating walking distance based on single accelerometer, Journal of Zhejiang University, Vol. 44, No. 9, 2010, pp. 1681-1686. [10] . A. J. Zhang, C. M. Wang, H. Zhao, Three axis electronic compass magnetic heading Angle error compensation research, Transducer and Microsystem, Vol. 27,No. 8,2008, pp. 33-35. [11] . Xu-Dong Yu, Yu Wang, Peng-Fei Zhang, Yuan-Ping Xie, Jian-Xun Tang, Xing-Wu Long, Calibration of RLG drift in single-axis rotation INS, Optics and Precision Engineering, Issue 6,2012, pp. 1201-1207. [12] . S. Moafipoor, D. A. Grejner-Brzezinska, C. K. Toth, Performance assessment of a multi-sensor personal navigator supported by an adaptive knowledge based system, in Proceedings of the 21st Congress of International Society for Photogrammetry and Remote Sensing, XXXVII, Part B5, 2008, pp. 857-866. Yu LIU, Wen-Ji XIONG Institute of Optical Communication Technology, Chongqing University of Posts and Telecommunications, Chongqing 400065, China Tel: 13658302450, fax: 021-62460380 E-mail: [email protected] Received: 10 September 2013 /Accepted: 25 October 2013 /Published: 30 November 2013 (c) 2013 International Frequency Sensor Association |