ION GNSS 2012
Session F3: MEMS Technology

Title: Robust Attitude Estimation for Indoor Pedestrian Navigation using MEMS Sensors
Author(s): M. Chowdhary, CSR Technology, USA; M. Sharma, A. Kumar, IIT, India; S. Dayal, CSR Technology, India; M. Kumar, IIT, India.
Date/Time: Thursday, September 20, 2012, 11:26 a.m.
Room: 206 (NCC)

Pedestrian Dead Reckoning (PDR) based on MEMS sensors have been used in hybrid systems to increase the accuracy, availability and robustness of location systems in recent times. Low cost MEMS sensors such as accelerometers and gyroscopes used in consumer applications, such as pedestrian navigation, are affected by sensor noise and drift which introduce errors in displacement and relative attitude (heading, roll and pitch) changes in the sensor´s frame of reference with respect to the human body (the "body frame"). In addition, a magnetic field sensor is used to provide an independent input for absolute heading with respect to Earth (the "earth frame"). The magnetic environment varies significantly in an indoor setting due to the presence of magnetic disturbances generated by various sources such as electrical equipments, computers, and mobile phones. Heading computations using only a magnetic sensor in such degraded environments will produce erroneous results unusable for pedestrian navigation. Attitude computation using MEMS sensors in a device such as mobile phone becomes even more difficult owing to the complex ambulatory motion over and above the motion of the center of mass of the human body with respect to the earth frame. These distortions are caused due to the effects of extra degrees of freedom and cannot be only attributed to the effective motion of the centre of mass in the body frame. In order to estimate velocity or compute heading using a compass, the attitude angles need to be determined in order to transform vectors from the body frame to the navigation frame of reference. These estimations can be achieved by using a fusion algorithm such as one based on Extended Kalman filtering. Pedestrian Dead-Reckoning algorithms that have no restriction on the location of sensor placement on the body increase the utility of the MEMS sensor based system such that they can be an integral part of any mobile consumer device. At the present time, most of the inertial measurement unit based positioning systems have constraints of locating the sensors either in the shoe, or near the waist. The algorithm presented in this paper for attitude estimation produces good results for arbitrary sensor placement on the body, and is resilient to extraneous motions like arm swing or trouser pocket. To take this fact in account, we have used a technique which adaptively changes the parameters of the measurement noise matrix R, and the system noise matrix Q in the extended Kalman filter formulation. In order to fuse acceleration and angular motion for computation of attitude, the R and Q matrices are made adaptive according to the user activity observed. Similarly, the fusion of magnetic sensor measurements with angular rate from gyro takes into account the magnetic disturbances observed. Since the position accuracy of a MEMS sensor based algorithms largely depends on correct estimation of attitude, this paper focuses on improving the estimation of attitude (roll, pitch and heading) in the presence of time-varying sensor drift, sensor noise, varying external magnetic field and dynamic movement of body parts. The accuracy of attitude and heading angles computation critically impact the performance of PDR mechanization, and thus increasing the accuracy for high dynamic conditions is the focus area of this paper.

Previous Abstract Return to Session F3 Next Abstract