動態環境駕駛輔助系統之多群組感測器選取演算法與軟體模擬

Abstract

現行的駕駛輔助系統是由獨立功能性的感測器所組合而成，用以觀測車子周圍的環境。這些駕駛輔助系統對於有些感測器有著逐漸增加的輔助需求，這些需求是有互補性的以及重疊性的。藉由融合這些感測器的資料，得到一個較大視野的探測環境，在一個適當的範圍內，量測的明確性及精確性是增加的。在不同的行車條件下，選取所需要的物體探測感測器是一個很重要的議題。在這篇論文中， 基於現行的駕駛輔助系統，我們提出了一個新型未來的駕駛輔助系統，藉由在不同的行車條件下的感測器種類及應用，在一個多群組感測器融合的架構下實行出來。在新型未來的駕駛輔助系統上裝設了在現行的駕駛輔助系統上通用的感測器。 在這篇論文中，發展了一個用於駕駛輔助系統的多群組感測器選取的演算法以及透過Borland C++ Builder軟體建立了一個模擬行車的介面。透過整數線性規劃及最佳化理論來描述用於新型未來的駕駛輔助系統的多群組感測器選取的問題。有兩個類型的議題用來考慮系統的實行成果：靜態（離線的感測器選取問題）；動態（線上的行車條件轉換問題）。我們提出了一個目標用來決定離線及線上的最佳化的感測器選取，需要考慮感測器的涵蓋區域、能量、頻寬及可靠度。基於多群組網路、不同的行車條件、行車安全性的轉換以及駕駛的需求，我們發展了一個用來探測行車環境及給予駕駛及時的警示的多群組感測器選取方法。而透過不同的模擬可以說明所提出策略的實行成果。

Fixed Driver Assistance Systems consist of functionally isolated sensors to observe the environment around vehicle. These driver assistance systems have increasing demands for several sensors, which are complementary but also redundant. By fusing these sensors data, a large field of view is obtained and the certainty and precision of the estimates in the relevant regions is increased. It is an important issue to choose necessary object-detecting sensors in different driving conditions. In this thesis, we present a new active Driver Assistance System which is based on fixed Driver Assistance Systems including sensor types and the applications for different driving conditions implements a multi-sensor fusion architecture. Commonly used sensors of fixed Driver Assistance Systems are deployed on the new active Driver Assistance System. In this thesis, a multi-sensor selection algorithm for Driver Assistance Systems is developed and a driving simulation interface is built up by Borland C++ Builder software. The multi-sensor selection problems for the new active Driver Assistance System is formulated by Integer Linear Programs (ILPs) and optimization theory. To consider system performances, there are two types of issues: static (off-line multi-sensor selection problem); dynamic (on-line driving condition transformation problem). We propose an objective to determine optimal selection of sensors off-line and on-line for guaranteed coverage, energy, bandwidth and reliability of sensors. Based on the multi-sensors network, different driving conditions, driving security transformation and driver commands, a multi-sensors selection approach for detecting the environment around the vehicle and warning the driver in time is developed. The performance of the proposed strategies is illustrated through extensive simulations.