Point Cloud Key Point Extraction Algorithm Based on Feature Space Value Filtering

With the rapid development of 3-dimensional (3D) acquisition technology, point clouds have a wide range application prospects in fields computer vision, autonomous driving, and robotics. Point cloud data is widely used many 3D scenes, deep learning has become mainstream research method for classification with advantages automatic feature extraction strong generalization ability. In this paper, hierarchical key framework proposed to solve problem modeling local geometric structure between points. Various models such as PointNet, PointNet++, DGCNN are analyzed their features extracted. Based on these analyses, an indexed edge spatial value screening neural network (IEGCNN) proposed. This extracts from each its neighborhood, calculates distance center points within adds orientation information network. The relationship architecture projected onto coordinate system decomposed into three orthogonal bases. two modeled by aggregation based angle vector base neighboring capability fast processing significantly reducing training recognition time. experimental results show that achieved high accuracy value. work also provides idea real-time target detection network, which broad applications prospect deployment movable devices processing.