Textured/textureless object recognition and pose estimation using RGB-D image

Abstract

In this paper, we propose a novel global object descriptor, so called Viewpoint oriented Color-Shape Histogram (VCSH), which combines 3D object's color and shape features. The descriptor is efficiently used in a real-time textured/textureless object recognition and 6D pose estimation system, while also applied for object localization in a coherent semantic map. We build the object model firstly by registering from multi-view color point clouds, and generate partial-view object color point clouds from different synthetic viewpoints. Thereafter, the extracted color and shape features are correlated as a VCSH to represent the corresponding object patch data. For object recognition, the object can be identified and its initial pose is estimated through matching within our built database. Afterwards the object pose can be optimized by utilizing an iterative closest point strategy. Therefore, all the objects in the observed area are finally recognized and their corresponding accurate poses are retrieved. We validate our approach through a large number of experiments, including daily complex scenarios and indoor semantic mapping. Our method is proven to be efficient by guaranteeing high object recognition rate, accurate pose estimation result as well as exhibiting the capability of dealing with environmental illumination changes.

Fast 6D pose estimation for texture-less objects from a single RGB image

“…Depth cameras (RGB-D) have been widely used by the scientific community with the recent development of low-cost devices, like Kinect. A number of algorithms have been proposed for computing 6D object pose from RGB or RGB-D data, or by combining both [7], [11], [9], [12], [13], [14] But when RGB-D sensors are not available, edge features from RGB data are typically used when the object texture does not provide enough information (i.e., it has low autocorrelation to compute proper, textured-based, descriptors). In this case, the pose can be estimated from correspondences between 2D line segments in the image and 3D edges in the model.…”

“…Template matching approaches [15], [16], [9], [17], [18], [13] have been successfully used with texture-less objects: these methods create a set of template images of the object from different viewpoints, and estimate the pose parameters by finding the template that is more similar to the observed image of the object. Different similarity measures have been proposed such as Chamfer distance [15], [16], image gradient orientations [19], point-to-point edge correspondences [18], object appearance [17], or by using custom descriptor-based distances [8], [9], [10], [13], [11].…”

“…Different similarity measures have been proposed such as Chamfer distance [15], [16], image gradient orientations [19], point-to-point edge correspondences [18], object appearance [17], or by using custom descriptor-based distances [8], [9], [10], [13], [11]. Note that the accuracy of template matching approaches heavily depends on the size of the template dataset: the larger it is, the closer to the actual pose are the estimated parameters.…”

Dense and Deformable Motion Extraction in Dynamic Scenes Based on Hierarchical MRF Optimization in RGB-D Images

“…In order to present the MRF node data comprehensively and efficiently, we design a new descriptor named Deformable Color-Shape Histogram (DCSH) which combines the shape and color features, as an extension work of [12].…”

“…According to the HSV value, each point p is mapped into two adjacent histogram regions RE u , RE u+1 in chromatic area, and one region RE 6 or RE 7 in achromatic area, with respective contributions being w u , w u+1 , w 6 |w 7 . For more details of color ranging process, see [12].…”