Object segmentation and recognition in 3D point cloud with language model

Yi Yang, Guang Yan, Hao Zhu, Mengyin Fu, Meiling Wang

Abstract: We proposed a recognition algorithm based on feature extraction and Bag-of-words with language model. Object recognition in 3D rang data of big scene has become an increasingly popular research topic in intelligent vehicle area. Firstly we introduce a fast ground filtering and clustering method and a convenient method to generation training samples. Then we proposed a feature extraction method based on 3D SIFT and FPFH to get descriptors with scale-invariant and rotation-invariant in 3D range data. Furthermore, we proposed a recognition algorithm based on bag-of-words and language model by add spatial semantic information to bag-of-words model to make up the shortcoming of ignoring the relationships between the visual words. At last, experimental results on real laser data depicting rural scenes are presented.