3D Mapping Considering Object Recognition Result Based on 3D LiDAR Information

This paper describes a 3D mapping method considering object recognition result by using 3D LiDAR sensor mounted on a wheeled mobile robot. The object recognition is realized by using a deep learning neural network called PointPillars. By comparing the recognition results between two continuous frames, the objects can be matched and for those which are recognized in both frames, and new ID will be given for new detected objects. The recognized objects with determined ID will be reflected to the 3D map.