| 摘要: | 自走車在應用上相對於人為駕駛有許多的潛在優勢，例如:可增加乘車空間；避免行車距離太短、駕駛者疲累等等因素造成的交通事故，而自動跟隨前導著軌跡的自走車，對於貨物的運送、區域的導覽都有相當多的應用。本研究藉由一個自主控制的系統，減少許多使用者的負擔；而本系統除了自走車本身外，僅需要一台電腦及PTZ相機與一塊arduino控制板，就可以完成跟隨前導者軌跡行進的動作。
論文架構主要分成三個部分，一是前導者的偵測及辨識、二是前導者的方位估計及追蹤、三是輪椅控制。前導者偵測是以前導者的梯度方向分佈圖 (histograms of oriented gradients, HOG) 做為特徵，經由支持向量機 (support vector machine, SVM) 分類判別是否為行人的物件，再將已經判定為行人的物件取色彩分佈圖來做為特徵辨識是否為前導者。第二部分的前導者方位估計，是透過相機取得前導者與自走車之間的相對座標，加上輪椅在世界座標中的位置，來取得前導者的世界座標。第三部份實作使用筆記型電腦搭載Intel® Pentium® CoreTM i7-5700HQ 2.70GHz中央處理器及伍氏科技的電動輪椅 Mambo 513和Logitech 的 QuickCam® Sphere AF PTZ 彩色相機和Arduino uno 控制板，透過電腦計算出輪椅該前進的距離傳送給 Arduino 控制板以電壓的方式輸出達到控制輪椅前進的效果。
在實驗分析中，我們拍攝校園內街道影片及實驗大樓大廳；透過行人偵測帶擷取垂直邊可達到99%將前導者取出誤判率約為49%，配合HOG特徵及SVM分類器偵測率約為85%誤判率約為0.1%。輪椅移動控制約可達到90%正確率，整體前導者軌跡重現平均誤差在20公分以內。系統在電腦上運算每秒約可以執行15張影像的處理速度。
;Self-propelled vehicles with respect to the applications of artificially driving have many potential advantages, such as: increase drive space; short driving distance to avoid traffic accidents, driver fatigue caused by, among other factors, and automatically follow the trajectory of the leading self-propelled car navigation for transportation, cargo area has a considerable number of applications. In this study, by an autonomous control system, reduce the number of user′s burden; and self-propelled vehicle in addition to the system itself, only need a computer with an arduino and PTZ camera control panel, you can follow the complete trajectory of predecessor action.
Paper architecture is divided into three parts, detect and identify one predecessor, the second is the leader′s direction estimation and tracking, three are wheelchair control. Predecessor gradient direction is detected by the leading distribution (histograms of oriented gradients, HOG) as a feature, via support vector machines (support vector machine, SVM) classification determines whether the object is a pedestrian, and then has been determined for pedestrians the object is to take color maps do feature recognition whether the predecessor. The second part of the predecessor bearing estimation, is to obtain the relative coordinates predecessor and between self-propelled vehicle through the camera, plus a wheelchair position in the world coordinates to get the world′s leading coordinate. The third part of the implementation using a laptop equipped with Intel® Pentium® CoreTM i7-5700HQ 2.70GHz CPU and Wu′s technology electric wheelchairs Mambo 513 and Logitech′s QuickCam® Sphere AF PTZ color camera and Arduino uno control panel, through the computer calculates the distance traveled wheelchair delivered to Arduino control board with voltage mode output to effect control the wheelchair forward.
In the experimental analysis, we shot the film on campus streets and laboratory building lobby; pedestrian detection by capturing with vertical sides can reach 99% of the predecessor remove false positive rate of about 49%, with HOG features and SVM classifier detection rate about 85% false positive rate of about 0.1%. Wheelchair movement control can reach about 90% accuracy, the overall trajectory reproduce predecessor average error of less than 20 cm. System on the computer can perform operations per second processing speed of about 15 images. |
