Pose estimation of a tilted pellet using single view from robot mounted camera Shraddha Chaudhary Department of Electrical Engineering Indian Institute of Technology Delhi New Delhi, India Chaudhary.shraddha18@gmail.com Ratan Sadanand Department of Mechanical Engineering Indian Institute of Technology Delhi New Delhi, India ratan.sadan@gmail.com Sumantra Dutta Roy Department of Electrical Engineering Indian Institute of Technology Delhi New Delhi, India sumantra@ee.iitd.ac.in Santanu Chaudhury Department of Electrical Engineering Indian Institute of Technology Delhi New Delhi, India schaudhury@gmail.com ABSTRACT This paper proposes a novel approach for finding the orientation of the tilted pellet using a single view of the pellet. This is important for the automation of pick and place problem. A texture less isolated cylindrical object is used for the pick and place operation. In this approach the isolated pellets are segmented from the background, following which, its contour is estimated. Then based on the assumption that length and diameter of the pellet remains constant, the mathematical formulation of the pose estimation of the pellet is given, and using which the orientation of the tilted pellet is computed using just single view. This is then experimentally verified for the different orientations of the pellet and the results are within the acceptable levels of accuracy A brief overview of the error estimation has been included. This error in orientation of the pellet is due to the variance in the actual height of the pellet. Error Jacobian for the above inaccuracy was calculated, and was found to be within required limits. Categories and Subject Descriptors I.4.9 [Image Processing and Computer Vision]: Applications. I.2.9 [Artificial Intelligence]: Robotics - Commercial robots and applications. I.4.8 [Image Processing and Computer Vision]: Scene analysis- Object recognition. J.6 [Computer Applications]: Computer-aided engineering - Computer-aided manufacturing (CAM). General Terms Measurement, Performance, Experimentation Keywords Computer vision, Robotics, Pose, Monocular vision, Curve based matching 1. INTRODUCTION In many manufacturing environments, work-pieces are supplied in plates. It is a common industrial problem to load machines with such pieces. Often, human labour is employed to load an automatic machine with work-pieces. Such jobs are monotonous and do not enrich human life. While the cost of labour is increasing, human performance remains essentially constant. Furthermore, the environment in manufacturing areas is generally unhealthy. Also, when work-pieces are inserted into machines by hands, there is a possibility of the limbs being exposed to danger. This is where the concept of process automation using computer vision plays a vital role. Use of computer vision in such cases is an obvious solution. Picking the work piece using a vision system would require the accurate estimation of its position and orientation (pose) using the data acquired by the vision sensor. Few of the known approaches for pose estimation include use of stereo vision [1] and range sensors [2]. Using the stereo set-up we can find the correspondence between the two images and thereby reconstruct the scene for further reference. However, generally objects used in industries are simple and featureless, which poses a problem to the researchers working in this area. In such a scenario, finding point correspondence for such objects becomes very difficult and even impossible in certain cases. In this paper we are proposing an approach that uses mono-vision model based recognition to solve 3D Pose estimation problem. The orientation of an isolated pellet was established from a single view of the camera, based on the known dimensions of the pellet and the workspace of the robot. But, the difficulty creeps in when we have to solve the 3D pose estimation problem using a single view, as 3D information of the object cannot be extracted directly from a single image. Since, we are taking just single view there is no alternative for establishing correspondence between two views and determine its change w.r.t to its horizontal plane Hence, in the image of the pellet looks like it is just oriented w.r.t to its vertical plane, therefore it becomes a non-trivial problem to solve for. 1.1 Related Work There has been a lot of research in the area of pose estimation of a 3D object using single view point [8, 9], in which using a single view important features are captured. This is advantageous in the aspect that a single view mitigates the problem of spatial correspondence between the views from the different cameras or multiple viewpoints. Similarly, [5, 6] have used monocular vision but in their work also focused on the look-up-table (LUT) scheme, through which the data was learned and hence the pose of the object was deciphered. But this scheme is very time consuming and not memory-efficient as lot of data has to be stored in the LUT for single pose of the pellet. This paper extends the work © 2015 Association for Computing Machinery. ACM acknowledges that this contribution was authored or co-authored by an employee, contractor or affiliate of a national government. As such, the Government retains a nonexclusive, royalty-free right to publish or reproduce this article, or to allow others to do so, for Government purposes only. AIR'15, July 02 - 04, 2015, Goa, India ©2015 ACM. ISBN 978-1-4503-3356-6/15/07…$15.00 DOI: http://dx.doi.org/10.1145/2783449.2783502