臺灣博碩士論文加值系統

無人載具近年來蓬勃發展，尤其是旋翼型無人飛行載具，許多空拍影像出現在各大媒體及網站，是現代科技討論熱烈的產業，世界各地爭相開發無人載具以提高作業效率。而在農業的應用中有自動噴灑農藥以及在空中航拍評估稻田所需的施肥量之應用，本論文整合完成一具備姿態感測、平衡控制、機器視覺辨識追隨以及障礙物形狀分類與迴避等能力，適用於低空果園及樹林木間穿梭之無人飛行載具。期望以機器學習為主要的控制模式，整備追蹤目標物與自主避障的功能。本研究使用大疆公司所生產的F450機架為主體，姿態計算採陀螺儀與加速度計取得的資料以互補濾波器進行姿態感測的數據融合演算，再將此數據以雙環式PID控制器進行馬達輸入的計算，以達到旋翼平衡；目標物辨識採機器視覺的方式進行識別，而雷射測距儀則用於取得目標物的距離以及對障礙物進行形狀分類；在運動控制上，載具會依據不同形狀的障礙物而有不同的避障策略。最終實驗結果顯示互補濾波器能有效改善陀螺儀的穩態誤差以及加速度計的高頻震盪；而在平衡控制上，雙環式PID控制上穩定性明顯優於單環式PID控制，然而在載具追隨與避障的實驗中，因顧及旋翼控制板的穩定性以及操作人員的安全，以地面式三輪全向無人載具進行測試。最終驗證結果顯示，本研究基於機器學習-倒傳遞類神經網路法所建構之載具系統已初步達成有效追隨目標、正確分類障礙物形狀以及賦予避障能力的研究目標。

Unmanned aerial vehicle (UAV) is flourish recently,; especially the quadcopter unmanned aerial vehicles. There are many aerial photography shown on the news and the websites and most of these photos are the contributions of the UVA. Since UAV is a hot topic of modern industry, manufacturers all over the world are devoted to developing the technology and application of UAV. In agriculture, many applications are conducted by UAV including self-spraying pesticide and evaluating the amount of fertilizer needed in field. This thesis develops an integrated system that consist of orientation sensing, balance control, machine vision recognition and shape sorting and avoidance of obstacle with which the UVA will meet the requirement of a low altitude shuttle in orchards and forest. This thesis also focus on employing machine learning as the main control core to achieve the performances of target following and obstacle avoidance for UAV. In order to maintain the balance of a quadcopter, the orientation data from gyroscope and accelerometer will be the inputs of a double loop PID controller. Target is recognized by machine vision. Distances and shapes of obstacles are defined by a laser rangefinder that mounted on the vehicle. The control of UAV, depending on the shape of the obstacle, execute different obstacle avoidance strategies. The results of experiment indicated that the complementary filter is better than gyroscope and accelerometer since it fixes steady-state error of gyroscope and oscillation in high frequency of accelerometer. Double-loop PID controller is more stable than single-loop PID controller is. Due to stability of control board and safety of operators in target following and obstacle avoidance experiment, a three wheels omni robot was employed to conduct the performance verification. The final results validate the novel UAV with capabilities of target following, obstacle sorting and obstacle avoidance equipped by the machine learning -the back propagation neural network.

目錄………………………………………………………………………v
表目錄…..………………………………………………………………..x
圖目錄…………………………………………………………………...xi
第一章 序論…………………………………………………………1
1.1前言…………………………………………………………………..1
1.2研究目的……………………………………………………………..1
第二章 文獻探討…..…………………………………………………3
2.1四旋翼的原理與基本介紹…………………………………………..3
2.2 姿態平衡控制……………………………………………………….4
2.2.1姿態量測原理…………………………………………………4
2.2.2 PID控制器……………………………………………………5
2.2.3姿態平衡控制文獻探討………………………………………6
2.3機器視覺文獻探討…………………………………………………..7
2.4 雷射測距儀的原理與特性………………………………………….9
2.4.1雷射測距文獻探討…………………………………………..10
2.5倒傳遞類神經網路…………………………………………………11
2.3.1 倒傳遞類神經網路文獻探討……………………………….12
第三章 實驗設備…..………………………………………………..14
3.1系統架構說明….…………………………………………………...14
3.2機體元件……………………………………………………………15
3.2.1電子調速器……………………………………………………….15
3.2.2無刷馬達………………………………………………………….17
3.2.3 F450四軸飛行機架………………………………………………18
3.3控制元件……………………………………………………………19
3.3.1 控制元件Raspberry Pi…………………………………………..19
3.3.2 控制元件Arduino Uno…………………………………………..21
3.3.3 GY-86十軸感測模組……………………………………………..24
3.3.4雷射測距儀………………………………………………………..28
3.3.5 Raspberry Pi Camera Module……………………………………..30
3.3.6 使用者圖形界面Qt………………………………………………31
第四章 控制流程設計………………………………..…………..…33
4.1 控制流程…...………………………………………………………33
4.2 姿態感測流程……………………………………………………...34
4.2.1姿態量測原理………………………………………………..34
4.2.2互補濾波器…………………………………………………..35
4.3平衡控制流程………………………………………………………35
4.3.1雙環式PID控制………………………………………………….36
4.4 機器視覺辨識流程………………………………………………...37
4.4.1 影像模糊化………………………………………………….38
4.4.2 HSV色彩空間轉換………………………………………….39
4.5基於機器學習的雷射測距避障策略………………………………40
4.5.1雷射測距資料前處理………………………………………..40
4.5.2倒傳遞類神經網路分類器…………………………………..41
4.5.2.1訓練模式……………………………………………………46
4.5.2.2回想模式……………………………………………………46
第五章 實驗方法……………………………………..…………..…47
5.1實驗方法……………………………………………………………47
5.2 姿態感測實驗……………………………………………………...47
5.3飛行載具平衡控制實驗……………………………………….…...47
5.3.1 Pitch與Roll軸PID控制實驗………………………….…...47
5.3.2旋翼飛行平衡PID控制實驗………………………...……...48
5.4機器視覺辨識實驗………………………………………………....49
5.5雷射測距儀辨識障礙物基於倒傳遞類神經分類器實驗…............49
5.6載具跟隨與避障實驗……………………………………….……...50
第六章 實驗結果與討論…………..………………………………..53
6.1姿態感測實驗結果與討論………………………………………....53
6.2飛行載具平衡控制實驗結果與討論……………………………....56
6.2.1 Pitch與Roll軸PID控制實驗結果與討論………….……...56
6.2.2 旋翼飛行平衡PID控制實驗結果與討論……………..…...58
6.3機器視覺辨識實驗結果與討論…………………………………....61
6.4雷射測距辨識障礙物基於倒傳遞類神經實驗結果與討論……....62
6.5載具跟隨與避障實驗結果………………………………….……...66
6.5.1無障礙物跟隨實驗結果與討論……………………………..66
6.5.2有障礙物跟隨實驗結果與討論……………………………..69
第七章 結論與建議…………………………………………………72
7.1結論…………………………………………………………………72
7.2建議…………………………………………………………………72
參考文獻………………………………………………………………..74

[1] 蔡伯萱。2014。”機器人基於模糊理論與基因演算法於目標追隨之研究”。台中：中興大學生物產業機電工程學系碩士論文。
[2] 簡貿謙。2016。”應用機器視覺於台灣五大水稻種子品種辨識之研究”。 台中：中興大學生物產業機電工程學系碩士論文。
[3] 羅茂榮。2014。”模糊理論應用於自主性機器人避障路徑導引之研究”。台中：中興大學生物產業機電工程學系碩士論文。
[4] 楊甄寧。2013。”基於人工勢場之自走車避障策略之實現”。新竹：清華大學生動力機械工程學系碩士論文。
[5] 黃國和。2006。”應用類神經網路與超音波感測器於車型機器人之路徑追蹤與避障”。台南：國立成功大學電機工程學系碩士論文。
[6] P. P. Rizky, .” Multi-copter development as a tool to determine the fertility of rice plants in the vegetation phase using aerial photos” Procedia Environmental Sciences, 2015, pp. 258-265.
[7] P. J. Benavidez. J. Lambert, A. Jaimes and, M. Jamshidi, .” Landing of a quadcopter on a mobile base using fuzzy logic, advance trends in soft computing,” Studies in Fuzziness and Soft Computing, 2014, pp. 429-437.
[8] L. E. Romero, , D. F. Pozo and , J. A. Rosales,. ”Quadcopter stabilization by using PID controllers,” Maskana I+D+ingenieria, 2014, pp. 175-186.
[9] P. Ryan. (2014, Aug 10).”Arduino Inverted Pendulum”[Online].
Available: http://rapot2014.blogspot.tw/2014/08/invertedpendulum3.html
[10] V. Mark. (2017, May 26). Using the force? no, it’s an apple watch flying this drone.[Online]. Available: https://pvdplus.wordpress.com
[11] K. J. Astrom and T. Hagglund,. “Process Models,” in PID Controllers: Theory, Design and Tuning, 2th 2nd ed. Instrument Societ of America. 1995. pp. 6-7.
[12] A. Tayebi and S. Mcgilvray, “Attitude Stabilization stabilization of a VTOL Quadrotor quadrotor Aircraftaircraft,” IEEE Transactions Trans. On Control Sys.tem Technol.ogy, vol. 14, no. 3, May 2006, pp. 562-57, May 2006.
[13] C. C. Peng and W. Singhose et al,.” using U sing machine vision and hand-motion control to improve crane operator performance,” IEEE transactions on Systems, Man and Cybernetics, Nov 2012, pp. 1496-1503
[14] R. Jain, R. Kasturi and B. G. Schunck. , “Machine vision.,” McGraw-Hill, 1995, pp. 1-2.
[15] U. Liz. (2014, Oct 22).” Real-time depth perception with the compute
module,”[Online]. Available: https://www.raspberrypi.org/blog/real-time-depth-perception-with-the-compute-module/
[16] F. Liu, S. Liang, X. Xian, H. Bi,. “Oscillation elimination for mobile robot based on behavior-memorizing,” ICCI Express Letters, 2011, pp. 3109-3115.
[17] R. Labayrade and D. Gruyer et al, .” Obstacle Detection Based on Fusion Between Stereovision and 2D Laser Scanner,” Mobile Robots: Perception & Navigation, 2007, pp. 91-109.
[18] S. B. Kotsiantis,. “Supervised machine learning: a review of classification techniques,” Informatica, July 2007. pp. 249-268.
[19] H. N. Chun. (2005). “Back-Propagation Network,” [Online]. Available: http://sjchen.im.nuu.edu.tw/MachineLearning/final/NN_BPN.pdf
[20] M. Lower and W. Tarnawski,. ”Quadrotor Navigation Using the PID and Neural Network Controller” Advances Intelligent Systems and Computing, July 2015, pp. 265-274.
[21] C. Zhang, H. Hu, J. Wang,.” An adaptive neural network approach to the tracking control of micro aerial vehicles in constrained space” International Journal of Systems Science, May 2015, pp. 84-94.
[22] D. Gandhi, L. Pinto and A. Gupta, . “Learning to fly by crashing”. The Robotics Institute,” Carnegie Mellon University, 2017.
[23] Hobbwing cop. (2016, Feb 10). “User Manual of Brushless Speed Controller“[Online]. Available: http://www.hobbywing.com/ products/enpdf/SkywalkerV2.pdf.
[24] Amazon. (2014).” A Pair MX2212 KV920 MARS POWER Outrunner Brushless Motor for DJI Phantom F450 F500 F550,” MX2212 data sheet.
[25] Raspberry Pi. (2016).” Raspberry Pi Compute Module 3,”Raspberry Pi 3 data sheet.
[26] Arduino. (2017). “Arduino UNO Rev3,” Arduino Uno data sheet.
[27] InvenSense, Inc. (2015, Feb 22). ” MPU-6000 and MPU-6050 Product Specification” [Online]. Available: https://www.invensense.com/wp-content /uploads/2015/02/MPU-6000-Datasheet1.pdf
[28] Measurement Specialties cop. (2012, Oct 25). “MS5611-01BA03 barometric pressure sensor” [Online]. Available:
http: //www.hpinfotech.ro/ MS5611-01BA03.pdf
[29] Hokuyo Automatic co., LTD. (2009, Aug 1). “Scanning laser range finder URG-04LX-UG01(Simple-URG)“ [Online]. Available: http:// www.hokuyo-aut.jp/02sensor/07scanner/download/pdf/URG-04LX_UG01_spec_en.pdf.
[30] Raspberry Pi. (2016).” Raspberry Pi Camera v2,” Raspberry Pi camera module data sheet.
[31] D. Hearn and M. P. Baker , (1986) Computer Graphics. Prentice Hall International, 1986, pp. 575-578.