Multi-Agent全天駕駛輔助系統

Abstract

駕駛輔助系統的發展是ITS的主要計畫之一，經過多年的發展以及許多方面的考量，包括環境、成本等，感測器與車輛的結合似乎是最佳的組合；單一的感測器雖能藉由所處理的訊息，加強某方面的安全性，但其感測的資訊有限，仍無法達到完整的輔助功能，於是，結合多種感測器，互相截長補短以強健輔助系統的能力將是較佳的解決方法。為此，本論文提出以多代理人架構整合之，利用代理人間的溝通特性達到感測器間互補不足的目的。 本論文之輔助系統是架構於多代理人系統上，以感測器為真實環境中的接收器，資料處理程式及多代理人系統是以VC++撰寫，並實現於PC系統上。目前，本系統具有CCD及雷射掃瞄器二個感測器，由CCD所接受的影像經處理後，將防偏、防撞訊息交給影像代理人作為警示依據，而由雷射掃瞄器所接收的距離資訊經處理後，將防撞訊息交給雷射掃瞄器代理人做為警示依據。由本論文列舉出的10個範例，可看出經由代理人間溝通後，可以解決單一感測器無法克服的問題。 本影像輔助系統已在高速公路上驗證其偵測能力，而其多代理人駕駛輔助系統，也已成功的架設於實驗用車(高爾夫球車)上，實現於交大校園環境中。

By combining the image sensor and the laser scanner, a driving assistance system is proposed in this thesis operated under different light and road conditions. Since the sensing data from all individual sensors are still limited to provide enough driving assistance functions, a feasible driving assistance system combining multiple sensors with exchanged information is proposed in this study to improve driving safety. Different types of sensors can be adopted and integrated as a multi-agent system so that communication with each other overcome the limitations of individual sensors and it leads to a more robust and capable driving assistance system. Two sensors are adopted in the proposed system, the CCD and the laser scanner. After processing the image data under different light conditions, the developed image agent obtains information of the road lanes and the front vehicles; different algorithms for day and night determined by an Otsu criterion are processed separately. Therefore, the functions of lane keeping and collision avoidance can be achieved for driving under all day. Moreover, after processing the distance data from the laser scanner, the laser-agent obtains all information of the front vehicles by applying the Kalman filter accurately. Finally, algorithms of all signal processing and multi-agent system are implemented using the VC++ . Ten typical models of driving conditions are discussed in this thesis and they indicate that through the communication of agents, difficult road conditions can be properly identified and the shortage of the single sensor is thus overcome. The detection capability of the imaging assistance system without laser radar has been proven on the high way; and furthermore, the multi-agent driving assistance system has also been realized on a golf car in the campus of National Chiao-Tung University.