Joint Detection of Articulated Vehicles Based on Region Growing for 3D Surfaces

10.29428/9789860544169.201801.0188

Chien-Choun Lin；Yu-Jyun Huang

point cloud ； bearing angle image ； 3D alignment ； region growing

NCS 2017 全國計算機會議

2017（2018 / 01 / 01）

1002 - 1006

英文

This paper proposes a novel joint detector approach for surfaces of articulated vehicles. The surfaces presented as 3D point clouds are captured by the laser scanner. The proposed approach converts 3D point cloud into 2D bearing images and then finds the corresponding pairs between two BA images by SURF. The corresponding 3D point candidates can also be obtained by the inverse mapping function of bearing image. Every candidate has an approximate normal vector derived by SVD with KNN and its nearest neighbor point. Thus, the rotation matrix between a corresponding point pair can be found. The points which are the best fit of SURF in BA images are considered as initial seed points of the adopted region growing method which criterion is the rotation matrices. In our experiments, the detection rate is 85.7% and the performance is efficient because the proposed method uses the features of 2D images to find the corresponding point pairs.