以環場影像作基於視覺的電腦輔助駕駛之研究

Abstract

本論文提出了一套基於電腦視覺及環場影像來作車輛輔助駕駛的系統，此系統具備在車輛行徑時偵測車輛移動狀態以及追蹤鄰近車輛的能力。我們設計了一個自動化的學習程序取代手動的操作來擷取影像中的特徵資訊。這些資訊包含了偵測車輛移動狀態以及追蹤鄰近車輛所需之資訊。我們提出一套偵測車輛移動狀態的方法透過在學習過程中取得之特徵資訊來分析輸入的影像並且識別出目前車輛的移動狀態。另外，我們設計了一套追蹤鄰近車輛的方法來偵測並且追蹤鄰近的車輛。接下來我們提出一套使用有限狀態自動機的狀態變化來偵測危險情況的方法來偵測一些可能發生的危險情況。我們最後以成功的實驗結果證明本系統的完整性與可行性。

A vision-based computer assisted driving system with the capabilities of analyzing vehicle movements and tracking neighboring vehicles using an omni-directional camera is proposed. A vehicle with computation and image grabbing capabilities is used as a test bed. An automatic learning process is designed for extraction of the features in grabbed images without manual instructions. Through this learning process, useful features for analyzing the vehicle movement and tracking neighboring vehicles are obtained. A vehicle movement analysis method is used to detect and analyze the current vehicle movement. Also, a vehicle tracking method is proposed to detect and track neighboring vehicles surround our vehicle. Furthermore, a risk condition detection method using a finite state-transition model is proposed to detect risk conditions in car driving. Finally, experimental results showing the flexibility of the proposed methods for computer-assisted driving in outdoor environment are also included. ii