Yönelim çıkarımı için arm tabanlı bir gömülü sistem tasarımı ve gerçeklenmesi
Design and implementation of an arm based embedded system for estimation of the orientation
- Tez No: 397802
- Danışmanlar: PROF. DR. MÜŞTAK ERHAN YALÇIN
- Tez Türü: Yüksek Lisans
- Konular: Elektrik ve Elektronik Mühendisliği, Electrical and Electronics Engineering
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 2015
- Dil: Türkçe
- Üniversite: İstanbul Teknik Üniversitesi
- Enstitü: Fen Bilimleri Enstitüsü
- Ana Bilim Dalı: Elektronik ve Haberleşme Mühendisliği Ana Bilim Dalı
- Bilim Dalı: Elektronik Mühendisliği Bilim Dalı
- Sayfa Sayısı: 107
Özet
Günümüzde geli¸sen teknolojiyle beraber dahili algılayıcı birimleriyle dı¸sarıdan gelebilecek gürültülerden az etkilenen, çevresel etkilere ve herhangi bir küresel koordinat sistemine ba˘glı olmadan, konum (bilinen bir noktadan) ve yönelim çıkarımları yapılabilmektedir. Kullanılan algılayıcılar; dönüölçer (gyroscope), ivmeölçer (acceloremeter) ve manyetometre (magnetometer) birimlerinden olu¸smaktadır. Her bir algılayıcı beraber kullanılarak, gidi¸s yönü ve referans sistemi (AHRS ya da MARG) bulunarak tam bir yönelim kestirimi yapılabilir. Bulunan yönelim yer çekimi ve Dünya'nın manyetik alanına ba˘glı tanımlanır. Bu tezde yapılan ça¸slı¸sma, ölçüm birimleri olan ivmeölçer, dönüölçer ve manyetometre kullanılarak kapalı alanda konum tahmini için ARM tabanlı bir gömülü sistem tasarımını ve gerçeklenmesini yapılmı¸stır. Kapalı alanda konum tahmini (Dead Reckoning) yapan sistem 32–bitlik ARM Cortex–M3 dü¸sük güç i¸slemci, 9 eksenli algılayıcı birimi, Bluetooth alıcı–verici, e¸s zamanlı çalı¸san iki ISM bandı alıcı–vericisi ile GNSS modülünü içeren geli¸stirme platformu olarak tasarlanmı¸stır. Adım belirleme, adım uzunlu˘gu tahmini, algılayıcı kalibrasyonu ve yönelim tahmini metotları kullanılarak kapalı alanda konum tahmini yapılmı¸stır ve yapılan deneyde sonuçlar verilmi¸stir. Bu çalı¸smada dahili algılayıcı birimleri; üç eksenli ivmeölçer, dönüölçer ve manyetometre kullanılarak yönelim çıkarımı algoritmaları incelenmi¸stir. Bu yönelim algoritmaları incelenmeden evvel ivmeölçer, dönüölçer ve manyetometreye ait matematiksel modeller verilerek fiziksel i¸slevleri anlatılmı¸stır. Yönelim gösterimlerinde kullanılacak koordinat sistemlerinin, Euler dönme açıları, dönme sırası, döndürme matrisi, dördey kümesinin temel matematiksel gösterimleri ve kendi aralarında yapılan dönü¸sümler için kullanılan e¸sitlikler verilmi¸stir. Yönelim ve algılayıcı toplama algoritmalarından Tamamlayıcı filtre (Complementary filtre) ve Kalman filtresi dördeyler yardımıyla anlatılmı¸stır. Her iki filtre içinde ivmeölçer ve manyetometreden bulunan yönelimler, dönüölçerin hatasını gidermek için kullanılmı¸stır. Ayrıca adım belirleme, adım uzunlu˘gu tahmini, algılayıcı kalibrasyonu ve kapalı alanda konum tahmini yöntemleri gösterilmi¸stir. Tasarlanan sistemin performansı, %2:2 mesafe hatası, %4 konum sapması ve %3;65 adım sayma hatası olarak ölçülmü¸stür. Bu hata oranları kısa mesafeler için makul seviyededir. Bu hatalar orta ve uzun mesafelerde, sistemin kümülatif hatalarından dolayı iyi sonuçlar vermeyecektir. Tasarlanan sistemin performansını artırmak için daha geli¸smi¸s kalibrasyon teknikleri, yönelim çıkarım algoritmaları kullanılmalıdır ve ba¸sarım oranı yüksek algılayıcı verisi toplama gereklidir.
Özet (Çeviri)
Recent developments in technology make it possible for integrated sensor units to measure the direction of movement without relying on any given global coordinate system as the reference. Three different sensor units are integrated into one package as the widely used integrated sensor units capable of evaluating such measure are gyroscope, accelerometer and magnetometer. By using three different sensor units all of which are designed to take measure from 3 different axis, the degree of freedom of the measure taken from such system is calculated as nine. Such system outcomes a full measure that creates the basis for estimating attitude and heading reference system such as the ones that relies on either AHRS or MARG. The orientation of the movement which is estimated from sensor units is directly related to the magnetic field of the earth and the gravity. The work presented in this thesis includes the design and implementation of an ARM based embedded system for PDR with MARG measurement units such three axes accelerometer, gyroscope and magnetometer. PDR system is based on 32–bit ARM Cortex–M3 low power processor and including 9–axes MEMS sensor unit, GNSS module with BLE and two concurrent ISM band transceivers is designed as a portable development platform. Step detection, step length estimation and orientation estimation methods are evaluated and combined as PDR model. We estimate the pedestrian walking position in a scenario sensor be placed on the hip. PDR system was tested and experimental results of its are shown. This work includes the explanation of the models of the inertial measurement units which are three axis accelerometer, gyroscope and magnetometer used to take measure in detail while supplying sufficient mathematical background to the reader to give an understanding about the applied filters to the sensor measures to extract more meaningful results. The Kalman Filter and the Complementary Filter are the filters applied to correct the estimated orientation using sensor measures. The mathematical backgrounds of the filters are explained using the Euler angles, sequence of rotation, rotation matrix and the quaternion representations. Furthermore sensor calibration, step detection, step length estimation and PDR system are evaluated. Given methodologies are explained in detail in the context of this thesis. A new board that is based on 32–bit low power processor and including 9–axis MEMS sensor unit, GNSS module with BLE and two concurrent ISM band transceivers is designed as a portable development platform. The details of the hardware configuration which consist of 32–bit ARM Cortex–M3 micro–controller, 9 axes MARG sensor unit, GNSS (supports GPS, GLONASS, GALIEO and BeiDou global satellite positioning services), wireless communication units which are 2:4 GHz BLE, 433MHz and 868MHz ISM band RF transceivers and RS422 serial communication converter for monitoring and debugging. Both control and connection of communication interfaces of peripheral unit around processor have been implied using STM32CubeMX which is owned by ST Microelectronics. Communication interfaces, initializing timers, and interrupt service routines can be adjusted by help of this program in the processor. STM32CubeMX program can automatically produce C programming codes of drivers which are at level of hardware for adjusted communication interfaces such I2C, SPI, UART, and interrupts and inputs–outputs. Furthermore time signals of the system can be adjusted making PLL adjustments which belong to units on the processor. Design and implementation of an ARM based embedded system which estimates the location of pedestrians with MARG sensors made with MEMS technology attached to body. The step detection, step length, attitude and heading reference can be estimated by collecting the data from MARG sensor unit. Before the estimation process the raw sensor data is calibrated and then The calibrated sensor data is used for orientation estimation. The peak detection algorithm is used as a step detection algorithm in this work. For the part of the orientation estimation, orientation is expressed with quaternion representation. Madgwick's orientation filter is used for the estimating the orientation and sensor fusion PDR system consists of sensor calibration, step detection, step length estimation and orientation estimation. Furthermore PDR model is explained in detail. Noise and discontinuous points exist on the data which is measure at the each axis for accelerometer, gyroscope, and magnetometer. Processing data in this manner causes errors while orientation estimation. Apparent affect builds up over time due to cumulative effect of consisted of errors. Calibration is necessary to minimize emergent errors. Biases are obtained using mean of the sensor data to calibrate data. Obtained biases are subtracted from raw data. Obtained values are filtered with second order low pass Butterworth filter. Both subtracting biases and filtering is implemented for other sensor units which are accelerometer and gyroscope. The first stage of walking distance estimation is step detection. There are several methods to detect number of step using accelerometer data. This step detection methods are peak detection method, flat zone detection method, zero crossing detection method and etc. Also x–axis data of gyroscope which is vertically located on the pocket is used in the some algorithm. In order to detect and count steps, both the peak detection and zero crossing detection are used via using accelerometer data. Peak detection and zero crossing detection have been used for detect the step in this algorithm in order to improve accuracy of step detection. When the differences of amplitude of peak values and amplitude of valley values are more than threshold and zero crossing occurs sequentially, this algorithm counts a step correctly. Threshold value is depending on walking speed. For the future it is possible to use adaptive system based on walking speed for updating the threshold. On the other hand, in that option increases computational load and our algorithm may be slower. The other stage of walking distance estimation is step length estimation. Step length estimation is based on different parameters such step frequency, obstacles and ground. Step length is a distance covered between the heel to heel during walking. Non–linear step length estimation is used to obtain the step length. As non–linear model has single parameter hence it is easy to determine and implement for real time applications. The orientation of any object can be obtained under favour of accelerometer, gyroscope and magnetometer sensors. Internal sensor units can perform measurements at the sensor frame system with accelerometer, gyroscope and magnetometer. Acceleration at the orientation of any object can be obtained using accelerometer both stationary acceleration (gravity) and linear acceleration which is originating from motion of object at three axes. Resultant angular velocity at each three axes is computed using gyroscope and then Earth's magnetic field is obtained measuring magnetic field at each three axes under favour of magnetometer. Two different orientation will be obtained using both gyroscope and accelerometer–magnetometer while orientation estimation. The orientation which will be used to compensate for obtained orientation errors using gyroscope, is obtained by accelerometer and magnetometer. One of the primary reasons of errors consisted of gyroscope is causing critical bias on the system in time even if they are stable due to their dynamic structure. Accelerometer and magnetometer are used to compensate for these errors. Obtained orientation using gyroscope is determinant while whirling. The filter which is used for the orientation estimation, has the same structure with implied filter by Madgwick. The filter provide making orientation estimation by combining two orientation estimations obtained with both gyroscope and accelerometer-magnetometers corresponding to their weights. So as to determine pedestrian movement, it should be known when the step occurs, step length and finally heading angle. With respect to these parameters, in a reference coordinate frame such as Cartesian coordinate system, the movement of pedestrian can be calculated with known initial condition. The heading angle is extracted from orientation filter in favor of accelerometer, gyroscope and magnetometer. Meanwhile step detection and length estimation are obtained from accelerometer. The results of each block can be combined and that makes possible determine the step vector. Finally all the step vectors gathered when step occurs form the trajectory. The evaluation methodology is determined as comparing PDR system with respect to ground truth. The performance of the PDR system was shown by position errors. The track walked during test was 18:7m x 7:35m square area. The experiment results were shown that estimated step length, estimated step counts and totally estimated distance walked were 0:671 m, 79 and 53:25m respectively. The performance of the PDR system is analyzed with experiment. According to the results of experiment is verified as %2:2 distance error, %4 maximum average positioning error and %3;65 step count error. The results are enough good for short range distances. On the other hand, for the medium and long range cumulative errors of PDR system will be increased. For the future work, performance of PDR systems should be enhanced and developed by advanced techniques for sensor calibration, increasing accuracy of orientation filters and sensor fusion algorithms.
Benzer Tezler
- Manipulation of visually recognized objects using deep learning
Görsel tanınan nesnelerin derin öğrenme kullanarak hareket ettirilmesi
ERTUĞRUL BAYRAKTAR
Doktora
İngilizce
2018
Mekatronik Mühendisliğiİstanbul Teknik ÜniversitesiMekatronik Mühendisliği Ana Bilim Dalı
DOÇ. DR. PINAR BOYRAZ
- Eş zamanlı konumlandırma ve haritalama tekniklerinin hız performansının geliştirilmesi
Improving runtime efficiency of simultaneous localization and mapping techniques
ZİYA UYGAR YENGİN
Yüksek Lisans
Türkçe
2017
Mekatronik Mühendisliğiİstanbul Teknik ÜniversitesiMekatronik Mühendisliği Ana Bilim Dalı
YRD. DOÇ. DR. VOLKAN SEZER
- Yapay zeka metotlarının bir sınıflandırma probleminde karşılaştırılması
Comparison of artificial intelligience methods for a classification problem
NİLGÜN ŞENGÖZ
Yüksek Lisans
Türkçe
2016
Endüstri ve Endüstri MühendisliğiSüleyman Demirel ÜniversitesiEndüstri Mühendisliği Ana Bilim Dalı
DOÇ. DR. GÜLTEKİN ÖZDEMİR
- Beyin bilgisayar arayüzü uygulamalarina yönelik EEG işaretleri için öznitelik çikarma
Feature extraction for EEG signals towards brain computer interface applications
ÖNDER AYDEMİR
Yüksek Lisans
Türkçe
2008
BiyomühendislikKaradeniz Teknik ÜniversitesiElektrik ve Elektronik Mühendisliği Bölümü
DOÇ. DR. TEMEL KAYIKÇIOĞLU
- Fotogrametrik dijital hava kameralarında kalibrasyon parametrelerinin zamansal değişim analizi
Analysis of the temporal change of photogrammetric aerial digital cameras calibration parameters
ABDULLAH KAYI
Doktora
Türkçe
2021
Jeodezi ve FotogrametriYıldız Teknik ÜniversitesiHarita Mühendisliği Ana Bilim Dalı
PROF. DR. BÜLENT BAYRAM