This paper describes an architecture designed to enable to detect an obstacle by monocular camera and create a path to avoid it. This technique consists of three parts: prefilter, occupied grid map, and octree-based A* algorithm. The prefilter includes initialization and update. As the initialization, four methods: Linear Least Square, inhomogeneous Singular Value Decomposition (SVD), homogeneous SVD, and nonlinear optimization, are compared. The recursive least square techinique is used for the update of the feature point position. This result is connected with the octree-based occupancy grid map. Finally, we inform the new A* algorithm based on the octree data structure.
展开▼