Multi-Agent 駕駛輔助系統

Abstract

本論文分為影像、雷射掃瞄器與系統整合三方面，在車道線影像方面，我們提出整合型車道線偵測方法，先利用IPM based之車道線偵測方法，找出車道線偵測視窗之定位，再由動態視窗進行車道線追跡，可精確定位視窗而減少影像處理的計算量，達到即時防偏偵測的需求。而在偵測前方車輛影像方面，我們則利用影像的對稱特性，能快速找出前方可能的位置，再進一步簡易的搜尋車輛下緣形狀特徵，以確認是否為前方車輛，達成快速的定位出前方車輛，達到防撞之功能。 在雷射掃瞄器方面，本文以一維的Kalman filter針對相對速度做估測，並利用相對距離與相對速度之間的關係，提出一 D/V 曲線碰撞預先警告時間。但此方法只針對物體以正向角度接近時才適用，因此，我們提出利用原點變換，建立六個虛擬的雷射掃瞄器，安排在雷射掃瞄器左右，利用此方法拓展監控範圍，使前方障礙物的偵測方法更加完備。 在駕駛輔助系統整合方面，依本論文應用來說，提出單獨使用影像可能會面臨的六個問題(Case 1 ~ Case 6)，及單獨使用雷射掃瞄器可能會面臨的兩個問題 (Case 7 ~ Case 8)，因此，本文最後引入多代理人溝通合作的觀念，透過自行開發的溝通管理伺服程式，來整合影像與雷射掃瞄器，並利用相互間資料交換，獲取所需的資訊，減少運算處理時間、降低不必要的錯誤或無法解決的問題，互補兩者間不足的地方，並由於溝通管理平台的建立，更容易增加其它感測代理程式，使得系統更具有彈性。最後，一個穩健的防偏與車前防撞駕駛輔助系統，在高爾夫球車上實現之。

The proposed system including three subsystems, the image, the laser scanner, and the multi-agent systems. In the image system, the lane detection is carried out by applying the inverse perspective mapping (IPM) to locate suitable windows for image processing to obtain the road lanes. Then, the on-line detection algorithms are processed to track the lanes efficiently. Also, the driving angle from the lanes are calculated. For detection of the front vehicles, the symmetrical characteristics of the vehicle image can be applied to locate the possible vehicle positions in the image. Furthermore, a simple searching approach can be applied to confirm the detection results. Thus, the lane keeping and collision avoidance can be achieved.

In the laser scanner subsystem, the estimation of relative speed between two cars is achieved by applying the 1-dimensional Kalman filter to obtain a reliable D/V curve. Thus, it provides the collision pre-warning time for the driver. However, it is not suitable for two cars not in the same line. A transformation of all measured signals by constructing six virtual lasers scanners is proposed in this thesis so that the present laser scanning system can be applied to a wide range.

To integrate both the image and the laser scanner as a driving assistance system, a multi-agent system is proposed to efficiently exchange information from two agents corresponding to the image and the laser scanner, separately. Six conditions are concluded that the image agent is not suitable for car detection alone; moreover, two cases indicate that the laser scanner may face difficulties in detecting the front car alone. Therefore, the multi-agent system is proposed in this thesis to integrate the image agent and the laser scanner agent. With mutual data and information exchange, the processing time is significantly reduced and the false detection is also surpressed. Furthermore, it provides the flexibility for future expansion with more sensors in this system. Finally, a multi-agent driving assistance system for the lane departure detection and the collision avoidance has been realized on a golf car tested successfully Sin the campus of National Chiao Tung University as well as the North Second High Way.