Hong Kong Applied Science and Technology Research Institute Company, Limited

The Chinese University of Hong Kong

Xueyan Tang

Suk Ling Li

Yunhui Liu

Meng Chen

Chun Li

Congyi Lv

Weilun Poon

Chongshan Liu

Hoi Fai Yu

You Ding

Computer implemented methods and computerized apparatus are provided for global registration between a first point cloud and a second point cloud obtained by a scanning device on an identical spatial scene at two separate instances. The method comprises extracting a first set of discriminative line-pairs from the first point cloud and a second set of discriminative line-pairs from the second point cloud, wherein a discriminative line-pair is a line-pair having high discriminative power compared to a randomly selected line-pair. In some embodiments, then a plurality of matching line-pair groups between the two sets of discriminative line-pairs are identified in accordance with one thresholding criterion related to between-line relationship, line geometry and line location; and a compass angle criterion related to compass errors of the scanning device. The method further comprises finding most reliable correspondence between the two point clouds by voting and then computing a global transformation matrix. Finally, the global transformation matrix is used to align the two point clouds. Embodiments of the present invention provide an accurate and efficient registration especially for building construction applications.

Method and system for global registration between 3D scans

Computer implemented methods and computerized apparatus for identifying planes of a 3D object in a 3D scene are provided. The method comprises receiving a point cloud and performs plane segmentation. In one embodiment, the point cloud is voxelized into a plurality of voxels of equal dimensions and the plurality of voxels are classified into three categories, wherein each first category voxel satisfies a planar requirement and a first neighborhood constraint, each second category voxel satisfies the planar requirement and a second neighborhood constraint, and each third category voxel does not satisfy the planar requirement. The method further comprises generating at least one reference plane from the first category voxels by first identifying a first category anchor voxel and recruiting its neighboring first category voxels to this at least one reference plane; and growing the at least one reference plane by absorbing second category voxels connecting to the at least one reference plane if they satisfy a voxel merging requirement and absorbing points in the third category voxels connecting to the at least one reference plane if they satisfy a point merging requirement. Then the set of at least one reference plane is used to represent the 3D object for further applications.

Method and apparatus for identifying planes of objects in 3D scenes

Computer implemented methods and computerized apparatus for posture, dimension and shape measurements of at least one 3D object in a scanned 3D scene are provided. The method comprises receiving a point cloud and performs 3D geometric feature extraction. In one embodiment, the 3D geometric feature extraction is based on a 3D hybrid voxel-point structure, which comprises a hybrid voxel-point based normal estimation, a hybrid voxel-point based plane segmentation, a voxel-based geometric filtering, a voxel-based edge detection and a hybrid voxel-point based line extraction. Through the process of 3D geometric feature extraction, the geometric features are then passed to the geometric-based dimension and shape measurements for various applications. After 3D geometric feature extraction, a further process of feature-based object alignment is performed. According to one embodiment of the present invention, using the exact lines extracted from the 3D geometric feature extraction, the computerized apparatus generates line-to-line (L2L) pair features for each identified object in the 3D scene. The L2L pair features in turn is used for aligning the identified 3D objects with target 3D objects.

Method and apparatus for posture, dimension and shape measurements of objects in 3D scenes

A system and method of detecting a posture of at least one predetermined part of an object comprising the steps of extracting a first object shape from a first image taken by a first image sensor, computing a characteristic dimension value of the predetermined part within a predetermined region of the first object shape, constructing a mask based on the characteristic dimension value, extracting a profile of the predetermined part from the first object shape by applying the mask on the first object shape, and identifying at least one predefined feature point in the profile of the predetermined part, thereby detecting the posture of the predetermined part is disclosed.

System and method for detecting and tracking multiple parts of an object

Disclosed are systems and methods for reproducing a body motion via a network, and systems, devices and methods for identifying a body motion. In an embodiment, the system for reproducing a body motion via a network comprises a sensor configured to capture a surface electromyography signal generated by a body motion of a user at a first terminal in the network; a processor configured to receive the signal from the sensor and identify the body motion based on the received signal, and send information associated with the body motion to a second terminal in the network, the processor being located in the first terminal; and a mechanical member configured to receive the information associated with the body motion from the second terminal and reproduce the body motion based on the received information.

Growing community of inventors

Hong Kong, China

Xueyan Tang