Doing this for each point in the sensed point cloud creates a mapping that associates each point in the sensed cloud to a point in the reference cloud. Nearest neighbor matching finds a candidate point in a reference point cloud that is nearest to a corresponding point in the sensed point cloud. We invite the reader to watch a video overview of the work detailed in this paper at.
For this reason, our research focuses on the Euclidean distance-based pairwise nearest neighbor matching employed by point set registration methods such as Besl’s ICP algorithm.
Our work has shown a nearly 90% speedup when compared to the traditional k-d tree as seen in Fig. After investigating the steps of ICP, when utilizing a k-d tree, the nearest neighbor step consumes the majority of the time, as shown in Fig. Thus, these applications require point-set registration processes be completed in a timely fashion. The system may determine an adjacent vehicle is veering into its lane, creating a potential hazard and requiring immediate action. In self-driving cars, on-board computers utilize the transformation found by ICP to detect objects within the car’s local environment. These point clouds can be known a priori, generated from images, light detection and ranging ( Lidar), or other sensors.Ī common point registration method, Iterative Closest Point (ICP), typically must execute in near real-time. Registration finds a rigid transformation between two or more data sets-providing a rotation and translation that minimizes the pairwise Euclidean difference between corresponding pairs of matched points. Point set registration presents a challenging problem for numerous applications including computer vision, pattern recognition, robotics, and image processing.
0 kommentar(er)
