在自走車的發展過程中,自動化動作的精準必須仰賴良好的運作系統,其中包含馬達的控制、路徑的規劃及障礙物偵測。本論文提出一套可供自走車偵測障礙物的感測系統,該系統具備障礙物的偵測及距離估測的能力。本論文提出一個基於視覺的感測方式,其特色是不需要處理環境複雜度高的立體視覺,而是以單一視覺感測器搭配前光源聚光型LED達到障礙物偵測及距離估測的目的。 在障礙物偵測的實驗上,透過影像的前處理擷取該LED的反射光源特徵。當只有一個反射光源特徵存在一段時間時,即判斷障礙物存在。以正確率、敏感度、有效性及信賴度對各種障礙物狀態的偵測能力進行系統的評估,其中信賴度是對系統判斷結果的可信任度評估,本系統為0.67,代表可信度良好。當障礙物有不規則的表面時,系統可克服單點及過度指向性的感測器(如,雷射及紅外線)偵測上的困難。 在距離估測方面,建立一個標準的實驗環境進行迴歸分析的方法與試誤法TaE (Trial and Error)的比較。由50個實驗數據以曲線擬合的方式得知距離估測方程式及參數。實驗得知迴歸分析的平均誤差百分比為-1%,TaE為-1.08%,該誤差比率小到足以符合多數自走車避障應用中的需求。
In the development of autonomous mobile vehicle (AMV), the precise automated actions must rely on a well-designed operating system, which includes motor control, path planning and obstacle detection. This thesis presents a sensing system for the obstacle detection of AMV. The system can detect the existence of an obstacle and estimate the distance between the AMV and the obstacle. This study presents a vision-based sensing approach, without the need of processing the highly complex environment of stereo imaging. Instead a single visual sensor with a directional front concentrating LED light is used to detect obstacles and estimate the distance. In the obstacle detection experiment, an image pre-processing step is adopted to capture the characteristics of a light reflection source obtained from the LED light. When only one light reflection characteristics is identified for a period of time, the system determines that the obstacle exists. The accuracy, sensitivity, specificity, and confidence level are used to evaluate the detection capability of the system for various obstacle conditions, where the confidence level reflects the trustworthiness of the proposed system. The proposed system got a confidence level of 0.67, which shows good credibility. When the obstacle has an irregular surface, the system can overcome the detection difficulty for a single pointing and over-directional sensor (e.g. laser and infrared sensor). In the distance estimation, a standard experiment environment is built for the comparative study between a regression analysis approach and TaE (Trial and Error). A curve fitting result from 50 experimental data is used to estimate the distance formula and parameters. The experiment shows that the regression analysis approach has -1 % of average error rate, while TaE has -1.08%. The error rate is small enough to meet the demand for the AMV in obstacle avoidance applications.