本文選題：ROV　切入點：姿態(tài)估計　出處：《哈爾濱工業(yè)大學(xué)》2017年博士論文　論文類型：學(xué)位論文

【摘要】：姿態(tài)估計具有廣泛的應(yīng)用,如空中,水下,機器人,導(dǎo)航系統(tǒng),游戲,工業(yè),增強現(xiàn)實系統(tǒng)等。目前,在該領(lǐng)域的深入研究已經(jīng)產(chǎn)生了許多完善的估計方法,其中復(fù)雜的如卡爾曼濾波,簡單的如互補濾波器。一般而言,傳統(tǒng)的姿態(tài)或角度估計濾波器的計算復(fù)雜度較高。為此,研究令人滿意的、精確的、計算復(fù)雜度低的算法是本文的初衷。因此,為了針對某些應(yīng)用,給出魯棒性強、簡單、高效的方法,互補濾波器(CF)得到了長足的發(fā)展。首先,互補濾波器的最新應(yīng)用是基于固定增益互補算法(FGCF)和漸變下降的互補算法(GDCF),該方法被用于基于微機電系統(tǒng)(MEMS)的慣性測量單元(IMU)中。這些固定增益估計器分別使用陀螺儀和加速度計進行高低頻姿態(tài)估計。結(jié)合不同的實際應(yīng)用,通過MPU6050 IMU的仿真和實驗驗證了GDCF和FGCF的性能。由于在沒有輔助傳感的情況下使用IMU,兩個濾波器的性能僅限于歐拉距離和側(cè)傾角度的姿態(tài)估計。兩者的估計結(jié)果相近,但是,FGCF比GDCF略有優(yōu)勢,其一是具有更高的精度,其二是該方法的兩個可調(diào)增益能夠提供額外的選擇。此外,相比于GDCF,FGCF濾波器增益的波動較小。兩種算法的計算復(fù)雜度幾乎相同。其次,本文分別使用FGCF和GDCF算法,以及擴展卡爾曼濾波法,進行MEMS IMU的姿態(tài)估計,并比較了估計的結(jié)果�；贛PU6050 IMU的仿真和實驗數(shù)據(jù),使用歐拉角度估計,對估計器的性能進行了評估,評估的依據(jù)是均方根誤差(RMSE)。此外,通過調(diào)整參數(shù)進行算法尋優(yōu)。結(jié)果表明,在不考慮計算負(fù)荷的前提下,卡爾曼濾波及其變體算法是解決位置和姿態(tài)估計問題的標(biāo)準(zhǔn)方法,FGCF和GDCF是解決此問題下的有效方法。結(jié)果評估中,EKF的效果最佳,但與CF相比,計算時間更長。與GDCF相比,FGCF有一點優(yōu)勢,部分原因在于FGCF的可調(diào)增益能提供更多的選擇。再次,FGCF、變增益互補濾波器(VGCF)和擴展卡爾曼濾波器(EKF)是許多應(yīng)用的有效解決方法,它們具有固定增益,計算復(fù)雜度分別為簡單、中等和復(fù)雜。MEMS IMU互補濾波器的精度,可以在少量計算的前提下,通過改變/切換濾波器增益的方法得到提高。這兩種方法都可以有效地用于輔助INS系統(tǒng),其中尋求較小計算負(fù)荷的算法是該應(yīng)用的主要研究方向。用于姿態(tài)估計的GDCF具有固定的增益,其數(shù)值不會隨系統(tǒng)的動態(tài)條件發(fā)生改變,這種情況會導(dǎo)致估計的錯誤。而復(fù)雜的算法由于具有較高的計算復(fù)雜度,不適用于大多數(shù)應(yīng)用對系統(tǒng)資源的限定。我們提出了模糊優(yōu)化互補濾波(FTCF)算法來消除誤差,并保證最小的計算負(fù)荷。所提出的算法與卡爾曼濾波算法進行了比較與評估。結(jié)果證明,與GDCF相比,FTCF大大減少了姿態(tài)估計的誤差。驗證了每個動態(tài)條件下濾波器增益的調(diào)整都在減小姿態(tài)估計誤差方面發(fā)揮了作用。此外,FTCF具有很小的計算成本,但其性能優(yōu)于GDCF,與復(fù)雜的卡爾曼濾波相近。最后,本文基于所提出的慣性SLAM算法,使用IMU的輸出數(shù)據(jù)和聲納觀察到的特征來估計潛水器的速度和姿態(tài),估計過程不使用其它諸如GPS等定位系統(tǒng)。慣性SLAM算法是INS和SLAM算法的組合。與EKF-SLAM相比,慣性SLAM的時間復(fù)雜度更低。所采用的粒子濾波器僅需使用較少的粒子數(shù)就可以達到EKF-SLAM的精度,并具有更快的計算速度。
[Abstract]:Attitude estimation has a wide range of applications, such as air, underwater robot, navigation system, game industry, augmented reality system. At present, research in the field has produced many perfect estimation methods, such as the complex Calman filter, as simple as complementary filter. In general, the calculation of complex filter the high degree of attitude or point of view of the traditional estimation. Therefore, research on satisfactory, accurate, low computational complexity of the algorithm is the original intention of this article. Therefore, in order to give some applications, strong robustness, simple method, high efficiency, complementary filter (CF) has got considerable development. Firstly, the complementary the new application of the filter is fixed gain complementary algorithm (FGCF) based on the gradient descent algorithm and complementary (GDCF), the method is used in microelectromechanical systems (MEMS) based on the inertial measurement unit (IMU). These fixed points gain estimator Don't use gyroscopes and accelerometers, high frequency attitude estimation. Combined with the practical application of different performance, GDCF and FGCF were verified by simulation and experiment of MPU6050 IMU. Due to the use of IMU in the absence of auxiliary sensor case, the performance of the two filters only Euler distance and inclination angle of the estimation results of attitude estimation. Similar, but FGCF GDCF than a slight advantage, one is with higher accuracy, the second is the method of the two adjustable gain can provide additional choices. In addition, compared to the GDCF, small fluctuation FGCF filter gain. Two algorithms for computing complexity is almost the same. Secondly, this paper uses FGCF and the GDCF algorithm, and the extended Calman filter method, MEMS IMU pose estimation, and compare the estimation results. The simulation and experimental data of MPU6050 based on IMU, using Euler angle estimation, the estimator The performance was evaluated on the basis of the assessment is the root mean square error (RMSE). In addition, algorithm optimization by adjusting the parameters. The results show that without considering the computational load, Calman filter and its variants of position and attitude estimation algorithm is used to solve problems of the standard method, FGCF and GDCF are the effective way to solve this problem the results in the assessment of the effect of EKF is best, but compared with CF, the computing time is longer. Compared with GDCF, FGCF has an advantage, in part because the FGCF adjustable gain can provide more choices. Once again, FGCF, variable gain complementary filter (VGCF) and the extended Calman filter (EKF) is effective to solve many application methods, they have a fixed gain, the computational complexity was simple, medium and complex.MEMS IMU complementary filter precision, can be a little calculation, by changing the switch / filter gain Improved. These two methods can be used effectively to assist INS system, which seek smaller computational load algorithm is the main research direction of the application. For attitude estimation GDCF with fixed gain, dynamic condition of its value is not changed with the system, this situation will lead to the error. And complex because the algorithm has high computational complexity, not suitable for most applications to system resources limited. We propose a fuzzy optimization complementary filtering (FTCF) algorithm to eliminate the error, and ensure the minimum computational load. The proposed algorithm and Calman filtering algorithm was compared with the assessment. The results showed that, compared with GDCF, FTCF can greatly reduce the errors of attitude estimation. Results show each dynamic condition filter to adjust the gain in reducing the attitude estimation error has played a role. In addition, FTCF has a very small The computational cost, but its performance is better than GDCF, similar to the Calman filter complex. Finally, the inertial SLAM algorithm based on the output data and sonar observed features using IMU to estimate the vehicle velocity and attitude estimation process, without the use of other such as GPS positioning system. Inertial SLAM algorithm is a combination of INS and the SLAM algorithm. Compared with EKF-SLAM, lower complexity of inertia SLAM time. The number of particles in particle filter only use less can reach the accuracy of EKF-SLAM, and has faster computing speed.

【學(xué)位授予單位】：哈爾濱工業(yè)大學(xué)
【學(xué)位級別】：博士
【學(xué)位授予年份】：2017
【分類號】：TN713;TN96
，

本文編號：1571841