自走車的自動化需仰賴良好的系統運作,其中包含障礙物偵測、路徑規劃及馬達控制等。在障礙物偵測上,有文獻提出光強度感應的方法,利用單一視覺感測器搭配一個聚光型LED手電筒,達到障礙物偵測與距離估測的目的。 在該方法中,以一連串的影像前處理步驟將LED的特徵光源擷取出來,而當只有一個特徵光源且持續一段時間時,代表已經偵測到障礙物。然而特徵光源擷取是否成功與障礙物的材質與型態有很大關係,使得該方法顯得不夠強健。根據該方法,本論文提出了修補破碎特徵光源的方法,並對新增的障礙物材質與型態的資料集進行實驗與分析,結果系統的信賴度由0.22大幅提升到0.64,證明本論文所提方法有效地改進了傳統光強度感應法的強健度與穩定性。在距離估測方面,本研究對40個實驗數據以迴歸分析的方式得知距離估測方程式及參數,並與雷射光影像測距為導向的IBDMS法以及IBDAMS法進行距離估測準確性的比較,最後列出IBDAMS法與本文所提方法之間的優缺點,以便讓使用者依自身的需求選擇其所需的視覺系統。本文所提方法的可用範圍為10 cm至200 cm以上,距離估測平均誤差為3.84%,此誤差比率小到足以符合多數自走車避障應用中的需求。最後將光強度障礙物偵測法加上簡單的避障策略植入自走車中,以避障實驗證明所提系統的實用價值。
The automation of autonomous mobile vehicle (AMV) must rely on a well-designed system operations, including obstacle detection, path planning and motor control. For obstacle detection, a previous work proposed a sensing method based on light intensity, where a single visual sensor along with a concentrated LED flashlight is used to achieve the objective of obstacle avoidance and distance estimation. In that method, a series of image pre-processing steps are adopted to capture the characteristics of reflection light obtained from the LED light source. Once the system identifies one and only one characteristic light source for a period of time, the system has detected an obstacle. However, it is not robust because the success of capturing characteristic light source highly depends on the material property and textures of the obstacle. According to that method, this thesis proposed a method to fix the potential problem of broken pieces of characteristic of the light source. By testing based on a new data set consisting of different materials and textures of obstacle, the proposed method is shown to be more robust and stable than the previous work because the statistic kappa value is pushed up significantly from 0.22 to 0.64. A regression analysis based on 40 experimental data is used to find a distance estimation equation and parameters in the distance estimation experiment. The proposed method and the laser image oriented methods called IBDMS and IBDAMS are compared in terms of the accuracy of distance estimation. The advantages and the disadvantages between IBDAMS and the proposed method are listed so that users can choose the visual system according to their needs. The applicable distance range with the proposed method is from 10 cm to more than 200 cm and the resulting absolute average error rate is 3.84%. The error rate is small enough to meet the requirement for the AMV in obstacle avoidance applications. Finally, the proposed method and a simple obstacle avoidance strategy is embedded in an AMV and a obstacle avoidance experiment is conducted to demonstrate the practical value of the proposed system.