移動軌跡記錄已泛用在各種領域上,例如醫學研究中的細胞移動、魚類追蹤等。生物學家能從生物的移動軌跡進行觀察,進而更瞭解生物的行為模式。目前運用電腦視覺技術紀錄物體的移動軌跡大多是使用平面影像,所紀錄的資料是物體在二維空間移動的座標;但在真實的世界中,二維空間座標並無法完整表達物體的移動軌跡。因此本研究製作了一套多物件的物件追蹤系統,透過立體視覺模擬人眼視角差的方法,記錄並顯示物體三維座標移動軌跡,使被追蹤物體的移動軌跡及速度能被更準確的計算出來。為了應用於真正的生物追蹤環境時方便架設實驗器材,本系統使用off-axis方式架設的兩部攝影機作為影像輸入,透過立體視覺模擬人眼視角差的方法之過程,記錄物體的三維移動軌跡,並利用顏色與深度資訊進行重疊物件區分。本系統的正確率可達95%。系統實做一3D軌跡展示子系統,方便使用者可以立刻觀察物件移動之軌跡。
Recording the object track is important for research in various fields, such as cell moving in medical research, fish behavior tracking, etc. Biologists can study the life behavior based on those records. Today, most of the object tracking are performed by analyzing the 2D coordinate data on the planar images. However, the 2D coordinate data can not truly present the object behaviors since most objects are moving in the three dimensional space. In order to track the 3D coordinate path, the tracking system must possess the ability of stereo vision to estimate depth of the object. This thesis tries to build a real time multiple object tracking system with the stereo vision to track the 3D trajectories of objects. Two conventional webcams are set up off-axis to mimic eyes of human beings. In addition, the colors of objects are incorporated as the auxiliary information. We test this system with the fish tracking problem under no fish-overlap and fish-overlap conditions. From the experiment results, in 10 frames per second speed, the correctness of tracking can achieve 95%. In order to show the trajectories of the objects, we also implement a subsystem to display the track in 3D manner.