Implementation and comparison of indoor location estimation methods on a handheld embedded system

Nowadays, Global Navigation Satellite Systems, which is called GNSS, has estimated the location of the person or vehicle with high accuracy. However, not exactly receiving of satallite signals or interruption of them has been observerd due to some disturbances. The studies about Personal Navigation Systems, which is called PNS, has recently been increasing to overcome these problems. Since the error of positioning increases cumulatively, PNS can not make accurately dead reckoninig whereas GNSS can do. Therefore, PNS can be used as complementary system. PNS devices can perform the estimation of the position of an object via detecting unit gathering data without depennding on geographical features and coordinate system of the world. This detecting unit is an electronic device and called Inertial Measurement Unit (IMU) consisting of three axis accelerometer, three axis gyroscope and three axis magnetometer. The integration of the electronic devices whose the sizes are reduced in micrometer into detecting device has been facilitated thanks to development of the technology of the micro electro mechanical sensor, which is called MEMS. Power consumption and cost of them decreases due to very small size of detecting unit. However, there are some disadvanteges. Measurement errors and electrical noise due to small size are the most disadvanteges. In this study, a device that can perform the position of an object as accurate as possible, overcome measurement errors and electircal noise of them, be easy to use in daily life and was designed. Different types of personal position estimating devices that can be placed on different location of human body are observed when researching recent studies about it. The most suitable and useful device among them is the one that can be handheld. Therefore, handheld devices are produced and various algorthims are used to solve the issues. Firstly, estimation of orientation was performed. There are Extended Kalman Filter, Linear Kalman Filter, Particle Filter and Madgwick Algorithm among algorithms estimating orientation in literature. In this study, it is proved that Madgwick Algorithm is better than Kalman Filters and Madgwick Algorithm was used. The estimation of orientation was performed by using quaternion after collecting data from 3 axis gyroscope, accelerometer and magnetometer. Secondly, it was explained how to mechanical and electrical noises on orientation angle during walking can be eliminated by using Kalman and Butterworth Filter.In this section, Kalman Filter and Butterworth filter were compared and it was showed that Kalman Filter was better than Butterworth filter because Kalman Filter could both estimate the next state vecor and filter the noises with noise variance. In addition to this section, an algorithm was proposed to prevent the error due to the hand deviation during walking. This algorithm is thought for labyrinth or corridor tracks. It is aimed that if the deviation angle is less than 5°, it will be round off the closest predefined eight special angles which are 0°, 45°, 90°, 135°, 180°, 225°, 270° and 315°. Finally, step detection and step length estimation algorithms were implemented. Several methods were proposed for step detection and an algorithm which detects the step by determining the maximum, minimum and zero points of 3 axis acceleration norm was used. Two filtersfrom literature, used for filtering the noises loaded on acceleration norm, were compared and best one was chosen. These filters are eight-step moving average filter and fourth-order low pass Butterworth filter with 2.5 Hz cut- off frequency. The next section is step length estimation after step detection. The most widely used method for step length estimation is the method proposed by Weinberg.In this method, since the step length coefficient is constant, it gives wrong results when walking in different step speeds. Adaptive step length estimation algorithms are studied in order to estimate step lengths at high accuracy in different step speeds and step lengths. Ho et al. also proposed a method of adaptively calculating the step length coefficient for handheld devices. In our work, this method was used to estimate the step length. A system was designed to implement the indoor location algorithms which were used in our work. This system consists of two units.The first one is the receiver unit and the other is the transmitter unit. The receiver unit provides the location mapping by sending the data to the laptop. This unit consists of an Arduino Uno card and a NRF24L01 2.4 GHz transreceiver.The data received by the receiving unit is sent to the laptop with serial communication at 115200 kb/s baud rate. The transmitting unit is the handheld device itself.This device sends the positioning data to the receiver after performing position estimation using the raw acceleration, angular velocity and magnetic field data from the intelligent IMU sensor. An intelligent LCDto display information , an intelligent IMU sensor, a NRF24L01 2.4 GHz transreceiver, and an Arduino Due card are on the transmitter unit. These raw datas from IMU sensor were processed with indoor location algorithms , the results are displayed on the LCD and they are sent to the receiver unit by Arduino Due, which has 32 bit Arm Cortex-M3 processor. After the system was designed, three different algorithms were tested at three different test fields and results were compared. Firstly, Madgwick Algorithm was used. Secondly, Madgwick+Conditional Orientation Algorithm was used. Lastly, Madgwick+Kalman+Conditional Orientation Algorithm was used. It was observed that best method is the algorithm combined with Madgwick, Kalman and Conditional Orientation algorithms. The performances of the algorithm combined with Madgwick, Kalman and Conditional Orientation algorithms used in three different test filed are %1.94 of distance error and % 0.364 of position deviation in rectengular test field, % 0.42 of distance error and % 0.34 of position deviation in labyrinthian test filed, % 0.331 distance error and % 2.017 position deviation in 45° inclined zig-zag test field. Recommendations that can be given from this work are that the preferred handheld position estimation devices due to the ease of use are more suitable if not used at the trajectories whose rotation angle is continuously changing. These trajectories cause the error accumulation. In addition, since the Kalman Filter runs slowly on the designed system, it must be designed specifically for applications where the orientation estimation needs to be calculated very quickly. Lastly, The hand should be released so that the handheld device can accelerate freely.