國立臺灣大學土木工程學系張堂賢邱偉承Chang, Tang-HsienTang-HsienChangChiu, Wei-ChengWei-ChengChiu2006-12-192018-07-092006-12-192018-07-092002-06http://ntur.lib.ntu.edu.tw//handle/246246/2006121215551141自動跟車的過程中，除了相對距離與相對速度的量測之外，對於前導車側向位移的量測，亦是不可或缺的控制變數。本研究以前導車目標影像灰階值的排列順序為特徵，由旋積值的運算結果得到影像匹配的相關係數值，當相關係數值大於事先設定的門檻值時則為辨識成功。本研究所建立的匹配模式最大匹配成功率為90%。當匹配失敗時，為保持前導車影像中心點座標資料的完整性，利用跟車中心點預測方式作為前導車影像之追蹤。經過實驗結果的比較發現，預測前導車影像的位置座標，以差值法最具有效性，但卡爾曼濾波器的預測則較具有實用性。為了節省處理時間，本研究提出 3σ 縮小搜索範圍區間，經實驗發現可提高7% 的運算速度。本研究綜理出之追蹤模式，更可利用其影像位移量經換算參數來推得前導車實際位移量，此樣本位移量可進一步作為推導道路曲率的參考資料。The variables in automatic car following involve not only relative distance and relative velocity, but also the lateral translation between the follower and lead vehicle. This study measured the translation of the lead vehicle track using real-time image processing. A convolution computation pattern matching method was used to track the lead vehicle image. The match rate success of the proposed method reached 90%. The predicted lead vehicle image location will replace the observed data while and if a pattern match fails. This study concluded that, in predicting the central image location of lead vehicle image, the extrapolation method is more efficient than the other methods. Reducing the matching region is proposed in this study to save processing time. A region defined within ±3σ from a previous scan can save 7% of the processing time. The tracking translation data can be the basis for estimating the roadway curvature during following.application/pdf423858 bytesapplication/pdfen-US自動跟車影像處理影像匹配車輛位移追蹤Automatic car followingImage processingImage matchingTranslation trackingjournal articlehttp://ntur.lib.ntu.edu.tw/bitstream/246246/2006121215551141/1/1160018424973415652355.pdf