Joint Object Class Sequencing and Trajectory Triangulation (JOST)

摘要: We introduce the problem of joint object class sequencing and trajectory triangulation (JOST), which is defined as reconstruction motion path a dynamic objects through scene from an unordered set images. leverage standard detection techinques to identify instances within registered Each these detections defines single 2D point with corresponding viewing ray. The rays attained aggregation all belonging common then used estimate denoted trajectory. Our method jointly determines topology reconstructs 3D points our detections. pose optimization over both unknown path, approximated by Generalized Minimum Spanning Tree (GMST) on multipartite graph refined continuous points. Experiments synthetic real datasets demonstrate effectiveness feasibility solve previously intractable problem.