近幾年隨著科技發展,可變式焦距透鏡系統廣泛的應用在日常生活中,然而透過使用可變焦攝影機將會產生嚴重的立體視覺問題,該問題為:倘若透鏡的焦距是不固定或非確切已知,則以傳統立體成像模型為基礎進行深度估算時,誤差將會隨之增加。為了改善上述的缺點,本文的研究目的為發展一部由雙攝影機構成之立體視覺影像系統,其係:第一,開發更精確的改良式雙目立體視覺演算法,以建立改良後的立體視覺模型,其係藉二個成對的立體影像來估算沒有焦距項的立體深度計算式,並藉由計算得出物體的真實世界座標;第二,經由改良式立體視覺演算法計算出目標物之物距,並將成果運用在立體視覺影像追蹤系統,且可運用在自主移動機器人上,以使機器人可以更精準的判斷障礙物進行避障並進行目標追蹤,而能夠更為迅速並且以高成功率完成任務。
With advances in technology in recent years, a variable focal length lens system widely used in daily life, however through the use of variable focus camera will produce a serious stereo vision problem, the question is: If the lens focal length is not fixed or exactly known, then a conventional three-dimensional imaging model as the basis for estimating the depth, the error will increase. In order to overcome the problem, the purpose of this study was the development of a constitution by a pair of stereoscopic video camera system, which system: first, the development of more accurate modified binocular stereo vision algorithms to establish improved stereoscopic model, the system by two pairs of stereo images to estimate the three-dimensional depth without focal entries formula and calculated by an object in the real world coordinates; secondly, through improved stereoscopic vision algorithms to calculate the object's object distance, and the results used in stereoscopic video tracking system, and can be used in autonomous mobile robot so that the robot can more accurately determine the obstacle avoidance and for target tracking, and can be more quickly and with a high success rate to complete the task.