臺灣博碩士論文加值系統

現今社會上有許多重度障礙者面臨長期無法出門與缺乏行動自主性之問題。雖然市面上目前有許多電動輪椅，提供搖桿、腦波以及聲控等不同的控制方式，但始終缺乏適用於重度脊髓損傷者之電動輪椅系統。對於重度障礙者，在操作輪椅時主要會面臨三種困境，首先是控制器的輸入方式，其次是電動輪椅之穩定性與安全性，最後則是電動輪椅之舒適性。本篇論文將提出一個結合類神經網路控制、光達障礙物偵測與多攝影機影像顯示與傳輸之電動輪椅駕駛輔助系統，藉此解決上述的第二個困境。
本篇論文所述之系統是由電動輪椅、嵌入式系統與平板電腦搭配光達、Wifi 網路、藍芽與編碼器等模組所構成。在系統架構上，為了未來因應不同使用者的需求採用模組化的設計，分為系統必要元件以及外加功能。
系統必要元件為電動輪椅、Android平板以及 STM32F4-Discovery控制器，使用者以 Android平板電腦做為操作介面，透過發展成熟的摩斯碼控制器輸入指令到平板電腦上，平板電腦再經由藍芽傳輸指令到負責控制輪椅的運行的 STM32F4-Discovery 控制器，藉此控制輪椅。由於傳統 PID 控制器在控制時需要根據不同環境以及使用者調整 PID 參數，因此在 STM32F4-Discovery上，加入類神經網路來學習輪椅系統的模型並調整PID的參數值，藉此增加電動輪椅在不同使用環境的適應能力。
外加功能部分可分為障礙物偵測、視覺死角影像顯示以及遠距離監控，可根據使用者的需求決定是否加入這些功能或是加入的數量。障礙物偵測功能主要使用 xv-11光達模組，可以掃描周圍360度的障礙物，由於在移動過程中可能發生本身量測的誤差以及運動狀態所產生的雜訊，因此使用卡爾曼濾波器對感測器資料做濾波，最後透過設計的演算法避開障礙物並調整輪椅的方向，也可在電動輪椅的前後各安裝一個。視覺死角影像顯示則採用 Rapberry Pi 連接兩個USB攝影機透過 Wifi 將影像傳輸到 Android 操作介面上，可顯示左右兩邊視覺死角的影像，如果有需要也可將此功能擴充到兩組，拍攝後方的影像。遠端監控功能則是透過網路將平板電腦與家中的電腦連線，藉此當輪椅上的加速度感測器判斷出異常的狀態時，可回傳警示訊息以及平板電腦鏡頭所拍攝到的路況影像。
實驗結果顯示，在輪椅控制方面，當類神經網路控制器訓練次數超過2次後，系統 PID 參數可達到收斂的效果，並且讓輪椅的移動達到穩定。實驗條件為在磁磚地板環境，並設定目標速度為每20ms行走10個編碼器脈波，當無載、有載但無人乘坐以及體重為75公斤使用者乘坐時，總共移動2000個編碼器脈波，左右輪之偏移距離皆為20個脈波以內。視覺死角影像顯示功能在160x120的解析度下可以達到 Android平板 每秒接收19.8幅影像。而遠距離監控功能在 3G 訊號良好且家用網路為20Mbs的情況下則可以達到每秒15.38福影像。
透過本論文所設計的系統，四肢癱瘓等重度障礙者可藉由嘴控、敲擊按鍵等能力輸入摩斯碼來操作電動輪椅。透過駕駛輔助裝置，可提升在操作上的安全性，遠端監控系統可在輪椅出狀況的時候即時傳遞訊息給照護者。

People with quadriplegia usually have less chance to go out because of the lost of mobility. Although there are a lot of commercial products providing different input methods such as joystick, brain wave, eye control and sound control to manipulate the electrical power wheelchair, none of them can fit their needs because of three main difficulties. First, in order to use the wheelchair for long time, the input device should be very comfortable. Moreover, high stability and safety of control system are absolutely required. Last, they have a huge blind spot during controlling the wheelchair because their vision is constraint by the lost of body mobility. The goal of this research is to overcome the second difficulties by applying neural network control, Lidar obstacle avoidance, multi-camera real-time blind spot image display and remote monitoring functions to the system.
In this research, we made a Smart Electrical Power Wheelchair (SEPW) system with embedded systems, Android tablet. Also, Lidar, Wifi, Bluetooth and encoder modules are also equipped to the wheelchair. In order to modify the system to fit different demands, we modularize the system and separate it to two parts – the main system and additional functions.
The main system part is including the electrical power wheelchair, Android tablet and STM32F4 Discovery controller. The user can use the well-developed Morse code controller to input command to the Android tablet. The command will be transmitted to the controller through Bluetooth. Then, the wheelchair executes the corresponding movement. In order to control the wheelchair, PID control method was applied to the previous research. However, due to the lack of adaptive ability, the PID parameter should be re-configured manually according to different user or environment. As a result, the neural network is applied and implement on STM32F4-Discovery to identify the system features and auto-configure the PID parameter, improving the adaptive ability of SEPW system.
Additional functions are obstacle avoidance, blind spot real-time image display and remote monitoring. These functions can be added to the main system according the different demands. The XV-11 Lidar sensor is applied to do obstacle avoidance. This sensor can perform 360 degree obstacle scanning and return the distance of obstacle in each degree. However, there are some errors caused by the motion or sensor. Kalman filter is applied to cancel these errors. After cancel the error, an obstacle avoidance algorithm presented in [2] was applied to navigate the wheelchair when approaching obstacles. Moreover, the number of Lidar can be increased depend on user’s demand. In order to display real-time image of blind spot, a Rapberry Pi connected with two USB cameras is installed to the wheelchair. The USB cameras can capture the image from left and right. Through Wifi, the images can be displayed on the corresponding position in the use interface of Android tablet. Similarly, more Rapberry Pi can be installed to the wheelchair to display more blind spot image. Last, the remote monitoring function is used to send warning message and real-time image to their family when the system detects the unusual motion pattern. This function is implement on the Android tablet and PC environment through internet socket programming.
The result shows that the PID parameter of the controller can approach optimal number when the neural network controller is trained after 2 commands. The wheelchair is move on the ceramic tile. The desire speed is set to 10 encoder pulses in 20ms. The input command is going for 2000 encoder pulse. The error between two wheels is lower than 20 pulse no matter the wheelchair is without load, moving on the ground or person sit on it. Real-time image of blind spot can be display with 19.8 frame per second when the image resolution is 160x120. Remote image monitor can reach 15.38 fps when Android tablet can receive adequate 3G signal and the PC with 20Mbps fiber Internet.
With this SEPW project, people with severe disabilities can control the SEPW safely, stably and comfortably. The remote monitor function also help the caregiver decrease anxious and worry.

摘要 II
Abstract IV
誌 謝 VII
Contents VIII
List of Figures XI
List of Tables XV
Chapter 1 Introduction 1
1.1 Motivation and Objective 1
1.2 Background 2
1.3 Related researches 4
1.3.1 Morse Code Controller 4
1.3.2 Smart Electrical Power Wheelchair Power Project 4
1.3.3 Self-Tuning PID Controller with Radial Basis Function Neural Network 5
1.3.4 Lidar Sensor for Obstacle Detection and Avoidance 6
Chapter 2 System Architecture and Design 8
2.1 Smart Electrical Power Wheelchair System Architecture 8
2.1.1 Basic Components 8
2.1.2 System Architecture 9
2.2 STM32F4-Discovery Embedded System Platform 12
2.2.1 STM32F4-Discovery 12
2.2.2 FreeRTOS 14
2.3 Rapberry Pi Embedded System Platform 17
2.3.1 Raspberry Pi 17
2.3.2 Difference between Raspberry Pi and Beagleboard-XM 19
2.4 XV-11 Lidar Module 20
2.4.1 Introduction to Lidar 20
2.4.2 Comparison of Lidar and Traditional Radar Sensor 21
Chapter 3 Methods 23
3.1 Neural Network Control 23
3.1.1 Incremental PID Controller 23
3.1.2 Radial Basis Function Neural Network 24
3.1.3 SEPW Control architecture 29
3.1.4 Software Flow Chart 32
3.1.5 Initial Output of Controller to Overcome Maximum Static Friction 33
3.2 Lidar Obstacle Detection 35
3.2.1 Kalman Filter 35
3.2.2 Obstacle Detection 39
3.3 Remote Monitor and Warning Message 40
3.3.1 Software Block Diagram 41
3.4 Real-Time Blind Spot Image Display 43
3.5 Software Integration 44
3.5.1 Task Scheduling of Controller 44
3.5.2 SEPW Android App 45
Chapter 4 Results 47
4.1 Neural Network PID Control of Electrical Power Wheelchair 47
4.1.1 Simulation Results 47
4.1.2 Experimental Environment and Details 49
4.1.3 Result without Load 52
4.1.4 Result with only the weight of wheelchair 54
4.1.5 Test result with weight of 75Kg person and wheelchair 56
4.1.6 Conclusion of Experiment Result 58
4.2 Lidar Obstacle Detection 59
4.2.1 Simulation Results of Kalman Filter 59
4.2.2 Practical Experiment Result of Kalman Filter 59
4.2.3 Obstacle Detection 62
4.3 Remote Monitoring and Control 66
4.4 Blind Spot Image Display 66
Chapter 5 Conclusion and Future work 67
5.1 Conclusion 67
5.2 Future work 67
References 68

[1]C. M. Wu and C. H. Luo. (2002). Morse Code Recognition System With Fuzzy Algorithm for Disabled Persons. J. Med. Eng. ＆ Tech., Vol. 26, Num. 5, pp. 202-207, 2002.
[2]C. H. Liang, C. M. Wu, S. W. Lin, and C. H. Luo. (2009). A Portable and Low-cost Assistive Computer Input Device for Quadriplegics. Technology and Disability, vol. 21, no. 3, 2009, pp. 67-78
[3]Ching-Hsing Luo, Pei-Feng Chen, and Li-Gang Young. (2012). Zigbee Morse Code Embedded System Design for Wireless Digital House Appliances. Proceedings of 2012 TREATS Annual Meeting, p6. June, 2012.
[4]J.Y Shiao, Y.C Wang, C.H Luo, P.C Chen, S.C Chen and M.D Shieh. (2015). Design and Approach of Interactive Assistive Electrical Power Wheelchair for Quadriplegic. Symposium on Engineering, Medicine and Biology Applications, Kaohsiung, Taiwan, Jan. 30 – Feb.1 2015.
[5]Wang Xiaoyuan, Fu Tao and Wang Xiaoguang. (2014). Design of radial basis function neural network controller for BLDC motor control system. Journal of Chemical and Pharmaceutical Research, 2014, 6(7):1076-1083
[6]Biao Yang, Jun Chang, Hezhou Su, Jinhui Peng , Shimin Zhang, Shenghui Guo, Libo Zhang, C. Srinivasakannan, Zilian Liu, Zhimin Li, Zhanyuan. Cao. (2013). Self-Adaptive PID Controller Integrated with RBFNN Identification Applied to Microwave Drying Process. Journal of Convergence Information Technology . Jan 2013, Vol. 8 Issue 1, p779-789. 11p.
[7]Hao Yu, Tiantian Xie, Stanisław Paszczyñski. (2012). Advantages of Radial Basis Function Networks for Dynamic System Design. IEEE Transactions on Industrial Electronics, Vol. 58, No. 12, December.
[8]Saurabh Ladha, Deepan Kishore Kumar, Pavitra Bhalla, Aditya Jain, R.K. Mittal. (2011). Use of LIDAR for Obstacle Avoidance by an Autonomous Aerial Vehicle. Third Symposium on Indoor Flight Issues, August, 2011, Grand Forks.
[9]Leandro C. Fernandes, Maurício A. Dias, Fernando S. Osório, and Denis F. Wolf. (2010). A Driving Assistance System for Navigation in Urban Environments. Lecture Notes in Computer Science Volume 6433, 2010, pp 542-551
[10]Keonyup Chu, Minchae Lee and Myoungho Sunwoo. (2012). Local Path Planning for Off-Road Autonomous Driving With Avoidance of Static Obstacles. Intelligent Transportation Systems, IEEE Transactions, Vol. 13, No. 4, December 2012.
[11]RoboPeak Team. (2014). RPLIDAR Low Cost 360 Degree 2D Laser Scanner (LIDAR) System Introduction and Datasheet. Rev.6, 2014 April 7.
[12]ST Semiconductor. (2014). UM1472 User manual - Discovery kit for STM32F407/417 lines. January 2014.
[13]Richard Barry. (2010). Using the FreeRTOS Real Time Kernel - A Practical Guide. 4461 Lulu.com. 2010
[14]John Moody, Christian J. Darken. (1989). Fast Learning in Networks of Locally-Tuned Processing Units. Neural Computation, Summer 1989, Vol. 1, No. 2, Pages 281-294