© 2013, IJARCSSE All Rights Reserved Page | 196 Volume 3, Issue 3, March 2013 ISSN: 2277 128X International Journal of Advanced Research in Computer Science and Software Engineering Research Paper Available online at: www.ijarcsse.com SIFT Based Approach: Object Recognition and Localization for Pick-and-Place System Girish Patil * Department of Electronics and Telecommunication, Govt. College of Engineering, Amravati, India Devendra Chaudhari Department of Electronics and Telecommunication, Govt. College of Engineering, Amravati, India Abstract — Vision based pick and place robotic systems have been the focus of significant research in both academia and industry. The system typically employs machine vision to analyse the scene, identify and locate the specified object and provide feedback to the robot arm for subsequent operations. For successful picking, the vision system needs to recognize the position and the orientation of the objects, the Scale Invariant Feature Transform (SIFT) is used for this purpose. The basis of the proposed work is built around two major areas; object recognition for developing artificial vision system and the robotics for carrying out the specified task with the specified object. In this paper, such object recognition techniques are reviewed. Keywords — Object Recognition, Scale Invariant Feature Transform, Image Processing, Feature Extraction, Robotics. I. INTRODUCTION The pick-and-place processes are the primary requisite for many of the industrial and household application. For such applications, there is a need to automate the pick-and-place process basically comprising of picking the intended objects, possibly performing certain tasks and placing them to desired location. The automated pick-and-place systems mainly consist of robotic arms and sensors. The machine vision is used as sensor and the primary function of them is to drive the robotic arms to the right location of desired object for picking and placing according to the robot’s degrees of freedom. The placing location is prefixed in most applications hence the sensors are rarely used here and the placing phase found to be comparatively easier. In contrast, the picking phase becomes very complex process in the applications where the scene is occluded and constrained. In this phase the sensors plays most important role as it is responsible for correct movements of the robotic system [1]. Most of the picking system assumes the situations where the objects are well structured, ordered, aligned and synchronized grasping of the objects. For such cases, the use of simple photocells will be sufficient to accomplish the picking phase. However, this approach will not be adequate for several applications as the arrangement to keep the objects well-structured and well aligned results in wastage of time and space of the process. In addition to this, there are some applications where the objects need to be kept in bins for saving time and/or for hygienic and safety reasons as shown in Fig.1. In this case, cameras used must be high resolution along with appropriate machine vision algorithms. In the scene, the objects are positioned at random inside a bin, a container or even at random on a belt/shelf; this problem is addressed as bin picking. [2] Fig. 1: Examples of complex situations for multiple object segmentation. The pick-and-place systems with robotic vision present several challenges, like the system should be capable of overcoming the difficulties in the disposal of the objects such as order, structure and placing of grasping points, working with every type of object of different dimension and complexity, with reflective surfaces or semi -transparent parts, such as in the case of pharmaceutical and cosmetic objects, often reflective or included in transparent flow packs, tackling the conditions of occlusion and clutter which make the object only partially visible, not only counting but also classifying the first instance of the object and also to identify all the duplicate objects with their orientation and dimensions, meeting the required working speed though having fast detection technique so that it works with several objects per minute.