本文主要是探討如何應用單一攝影機實現立體影像偵測技術,並應用於路邊停車輔助系統之停車空間估測。移動立體視覺是以單顆攝影機擷取影像,取樣時間前後兩張影像為一組,利用後軸兩輪之輪速感知器得到相對之距離與角度關係,進行立體視覺之深度計算。本文利用邊緣偵測搜尋一組影像上之特徵點,接著使用誤差絕對值總和與條件限制,減少特徵點配對誤差,並以同軸幾何演算法得到環境深度資訊。本文並利用車後單一攝影機與車側單一攝影機進行感知器融合,將其個別得到之深度資訊進行疊合,估測出停車空間之環境地圖,最後利用雷射雷達驗證估測之結果。
This research investigates how to employ single camera to for stereo vision detection for parking space detection of the parking assistance system. Motion stereo employs single camera to capture image. Images captured between the sample time are specified as a pair. The relative distance and orientation obtained from odometry are used to calculate the depth information using the stereo vision approach. Edge detection is used to find the feature points of two images in a pair. Sum of absolute difference and restriction rules are used to reduce the feature matching error. Epipolar geometry is then used to obtain the depth information of the environment. Two cameras installed at the rear-end center position and right side of the vehicle are used for sensor fusion. Depth information obtained from these two cameras are overlapped to estimate the environment map of the parking space. Laser scanner is used to verify the estimation result.