Robotic Grasping Based on Efficient Tracking and Visual Servoing using Local Feature Descriptors

In service robotic applications, grasping daily objects is an essential requirement. In this context, object and obstacle detection are used to find the desired object and to plan an obstacle-free path for a robot to successfully manipulate the object. In this paper, we propose a high-speed object tracking method based on a window approach and a local feature descriptor called speeded-up robust features (SURF). Instead of tracking the object in full image, we search and match features in the window of attention that contains only the object. Therefore, the tracked interest points are more repeatable and robust against noise. The visual servo controller uses geometrical features that are computed directly from the set of interest points, which makes the method robust against the loss of features caused by occlusion or changes in the viewpoint. Furthermore, these features decouple the translations and rotations from the image Jacobian, and also keep the object inside the camera's field of view. Various experiments with a robotic arm equipped with a monocular eye-in-hand camera demonstrate that objects can be grasped safely and in a stable manner in a cluttered environment using the proposed method.