运用多视图几何原理重建鱼类游泳三维轨迹

鱼类行为学数据是水质污染监测的主要依据,也是鱼类自身健康状况的反应,提取鱼类行为学数据对水质监测研究至关重要。为了获取鱼类行为学数据,实验提出了一种运用多视图几何原理研究鱼类轨迹跟踪重建的方法,实验中用3个摄像头分别从正面、侧面、俯视面3个方向监控水箱中鱼的游动,通过前景跟踪法在3个摄像头中分别得到一条二维轨迹,再利用多视图几何融合,计算出鱼类在水箱中游动的三维轨迹,从而可以重建鱼类的游动轨迹。实验重建了单条鱼在水箱中游动的三维轨迹,计算出了鱼类的位置信息和实时速度。研究表明,重建轨迹与鱼的真实轨迹虽然存在0.88~2.88 s的延迟,但结果吻合度较高。

Fish swimming 3D trajectory reconstruction based on multi-view geometry

Fish behavior data are quite important to water pollution monitoring and these kinds of data sets imply the fish healthy conditions. To acquire fish behavior data, we present a multiple-view based fish swimming 3D trajectory reconstruction technique. In this technique, we use three cameras to observe an aquarium from three different aspects and employ foreground detection and tracking methods to reconstruct 2D fish swimming trajectories from these three cameras. Based on the multiple-view 3D reconstruction theories, we fuse these three 2D fish swimming trajectories to achieve a 3D fish swimming trajectory. Experiments show that our method can accurately reconstruct the 3D swimming trajectory of a single fish in a cubic aquarium and get the real-time 3D position and the swimming speed of the fish. This method needs neither expensive experimental equipment nor strict installation requirement to make cameras vertically face the aquarium faces. However, it can acquire accurate 3D fish swimming trajectories in real-time.