Typical Mobility Reconstruction For Rigid Objects Based On RGB-D Camera

With the development of 3D scanner,it is possible to acquire 3D points on surface of objects in real world.Particularly,it is cheaper to capture these 3D data after some depth cameras appeared,such as Microsoft Kinect,Asus Xtion,Intel RealSense and so on.Nowadays,the 3D scanners can capture 3D points on surface of various objects both man-made and natural,so many 3D geometry reconstruction technologies to model objects in real world have be proposed.However,the information that static object model has is not enough in some degree.Therefore,more and more researchers start to discover new scanning analysis technologies to extract dynamic information of dynamic objects,for example,reconstructing mobility.Compared with geometry,mobility encodes the functional information of objects more definitely.In fact,mobility of object includes much valuable information.As for an articulated object,the mobility is composed of attributes of each joint,including joint type,joint position,joint pose,kinematic velocity and other space-time parameters.These attributes are very important in fields of mechanical interactions,robot dynamics,human body motions,computer vision and so on.Nevertheless,it is a challenging task to analyse mobility of articulated objects by 3D dynamic data captured by 3D scanner,which is often caused by the low quality of the scanned 3D dynamic data.3D scanners are limited by resolution and frame rate.When applied to rapidly moving geometries,the resulting point sets are typically noisy and sparse.Furthermore,moving parts on objects may generate significant outliers due to ghost effects and self-occlusions yielding absence of the data.This paper proposed a 4D RANSAC algorithm to extract mobility of rigid articulated objects from low-quality dynamic data captured by depth camera directly.Different from traditional mobility reconstruction method,we directly extract motion type,motion parameters,freedom for each joint without geometry reconstruction.The advantage of RANSAC lies in its robustness.Although data may consist of large outliers,it can accurately estimate the parameters of the model that optimally fits this data.Similar to traditional RANSAC,the method in this paper searches for the best mobility that fit to the 4D spatio-temporal dataset.There is three typical mobility in real world:Hinge,Ball Joint and Slider.So in this paper,we just take account of these mobility.However,the method proposed in this paper can absolutely handle other kinds of mobility.Furthermore,this paper proposed mobility graph to represent the articulated objects.This can be regarded as a method about skeleton reconstruction for objects,which can be used for object recognition and retrieving.