Object Detection in Non-Structured Environments with Application to Robot Grasping

C. Fernández and M.A. Vicente (Spain)


Snakes, potential fields, object recognition, robot grasping.


An algorithm for object detection in robot grasping appli cations is presented. The goal is to detect the object to be grasped with enough precision as to be able to com pute its optimum grasping points. The initial detection is carried out using an appearance based approach, which gi ves a first, rough segmentation. In a second step, defor mable contours or snakes are used to adapt the initial seg mentation to the borders present in the image. The main contribution of the paper is the proposal of different poten tial fields which, unlike traditional potential fields based on the gradient of blurred border images, work properly even if the previous segmentation process is not completely pre cise and the initial snake is located away from the image borders. Three different potential functions are are compa red, and the experimental results show that a novel, smooth function outperforms traditional potential functions both in terms of segmentation quality and speed of convergence. Some robot grasp examples obtained using our algorithm are shown in order to confirm the validity of the approach.

Important Links:

Go Back