無人車雙眼視覺避障系統之設計

Obstacle Avoidance System Design of An Unmanned Vehicle with Stereo Vision

單幼玫(Yu-May Shan)

指導教授 ： 王立昇

共同指導教授 ： 張帆人(Fan-Ren Chang)

國立臺灣大學/工學院/應用力學研究所/碩士(2009年)

https://doi.org/10.6342/NTU.2009.02845

摘要

本研究主旨為發展一戶外自動導航車系統。系統包含主控站、參考站和無人車三個部份。無人車整合了GPS接收機、電子羅盤、雙眼相機和六個超音波感測器等感測元件。無人車預先偵測前方障礙物形狀與位置，且在障礙物進入其運動方向之安全範圍時進行避障；藉由全球定位系統載波相位三次差分法（KGPS）進行全域定位，作為在無障礙物之狀況下之控制資訊；方位角則由電子羅盤輸出獲得，經由無線網路構成的各次系統資訊交換機制，使主控站得以監控無人車行為。
雙眼視覺處理方面，利用左右影像疊合時會產生一平面視差之原理得到無人車前方地理資訊。首先利用邊緣偵測取出左右影像輪廓，再使用影像處理方法將影像疊合找出並標出障礙物與地面的交界線。最後經由同點異像矩陣計算出前方障礙物的位置與距離，作為避障資訊之參考。
地形避障控制方面，引入位能場概念。由車體至終點及前方障礙物的距離分別計算引力及斥力；依障礙物邊緣座標點設計高斯分佈之能量場，計算總位能；其後使無人車行經於最小位能處所形成之平滑曲線，並控制其角度和速度輸出以規劃路徑。
即時避障控制方面，利用VHDL語言配合FPGA開發板設計一控制六顆超音波感測器輸出輸入電路，再透過超音波感測器得到距離資訊。此六個超音波感測器乃由設計控制器管理，經由規則庫系統推論無人車所需要的角度和速度輸出，對移動中的障礙物進行即時避障控制。
將上述感測器及其避障模式加以整合，使無人車在不同環境中能避障及抵達終點。觀察各個避障模式結果，驗證本論文的理論確實可行。

關鍵字

雙眼相機 ； 立體視覺 ； 平面視差法 ； 位能場模型法 ； 導航避障控制

並列摘要

The main theme of this thesis is to develop an outdoor navigation vehicle system. This system consists of three parts which are the main station, reference station, and an unmanned vehicle. The GPS receiver, electronic compass, stereo camera, and six ultrasonic sensors are integrated into the unmanned vehicle. The obstacle avoidance is performed by the vehicle which can detect the shape and position of the obstacles within a detectable range in the direction of its motion. The position of the unmanned vehicle is provided by GPS in an obstacle-free situation with the help of Triple Difference Carrier Phase method KGPS. The Heading angle is obtained through the output of an electronic compass. In addition, the main station can monitor the performance and behavior of the vehicle via the data exchange mechanism constructed by several sub-systems and the wireless network.
With stereo vision, the geographic information ahead of the vehicle is obtained by the principle of plane induced parallax. The profiles of left and right vision are acquired by Canny Edge Detection algorithm. Then, an intersection line of the obstacle and the level ground is found by the way of image processing. Finally, the position and distance of the obstacle is calculated by the homography matrix method.
The method of potential field is used to perform obstacles avoidance. The distance between the vehicle and the destination is used to compute the attractive force and those between the vehicle and obstacles are used to find the repulsive force. Gaussian distribution-like potential energy is derived for each obstacle. Then, the unmanned vehicle is required to move on a smooth curve with lowest potential energy by controlling its angular velocity and the motion velocity.
For real time obstacle avoidance, the input and output circuits of the six ultrasonic sensors are designed by VHDL language and implemented by FPGA development kit. The information of distance is obtained through the ultrasonic sensors which are managed by the design controller. Finally, the real time moving obstacle avoidance is performed by the angle and velocity that are inferred from the rule-based system.
Integrating the above-distributed schemes unmanned vehicle can avoid collisions with obstacles and arrive at the destination in different scenario. Experimental results show that the proposed navigation and control methodology are practicable.

並列關鍵字

Stereo camera ； Stereo vision ； Plane induced parallax ； Potential field model method ； Navigation obstacle control

參考文獻

[16]吳宣誼，“雙無人載具協同控制與實驗”，台灣大學應用力學研究所碩士論文, 中華民國九十四年七月。

[17]吳中遠，“即時視覺輔助戶外無人車系統設計與控制”，台灣大學應用力學研究所碩士論文, 中華民國九十五年七月。

[18]蘇信豪，”無人載具避障之導航與控制”，台灣大學應用力學研究所碩士論文, 中華民國九十六年七月。

[1]A. Ramírez-Serrano and M. Boumedine, “Ultrasonic Sensing and Fuzzy Logic Control for Navigation in Unknown Static Environments”, IEEE Advanced Mobile Robot, Oct., pp.54-59, 1996.

[5]G. Ramírez, and S. Zeghloul, “Collision-free path planning for nonholonomic robots using a new obstacle representation in the velocity space”, Robotica, Vol. 19, pp.543-555, 2001.