International Journal of Computer Science Trends and Technology (IJCST) – Volume 3 Issue 2, Mar-Apr 2015 ISSN: 2347-8578 www.ijcstjournal.org Page 88 Object Detection and Localization Using SURF Supported By K-NN Upendra Singh [1] , Sidhant Shekhar Singh [2] , Manish Kumar Srivastava [3] Research Scholar [1], [2] & [3] Department of Computer Science and Engineering Madan Mohan Malaviya University of Technology Gorakhpur - 273010 UP - India I. INTRODUCTION In the area of intelligent systems, object detection which is the task for searching and localizing objects in a scene is deliberated as prime feature for autonomy. This fact has impelled unprecedented research in this field and as a result several algorithms have been proposed in last two decades. The performance of all these algorithms has been promising up to some extent but the global acceptance of a single algorithm is still debatable. It has been frequently regarded that human visual system works on the pattern in which it makes series of fixations at various conspicuous locations in an image while observing static or dynamic scene. It has spurred many researchers but unfortunately very little is known about operations carried out in human eye while fixation. In our proposed method, we postulate that during a fixation, before recognizing any object human eye first of all segments that object out of that image in lieu of segmenting the entire image at once. So, we perform region based segmentation with the help of fixated points. For detection task, literature witnesses many algorithms have been suggested. All previous proposed algorithms share a common problem that they succumb in case of limited image information. Limited information can be understood as the diminutive size of object and occlusion. To tackle these problems, Mae et al. resorted on a local feature matching algorithm which uses local geometry consistency for detection task. This approach uses SIFT for feature extraction and matches with those of reference image. Despite the advantage of simplicity, it does bring some limitations i.e. non-planner surfaces pose an adverse effect on performance. So, the field of object detection still requires further research in order to achieve 100% accuracy. To meet the expectations and initiate further research in this field, this paper proposes an object detection algorithm based SURF algorithm assisted by fixation based segmentation. Dealing with real-life situations, due to perspective changes, various types of geometrical deformations are introduced in our image. The ABSTRACT The ability to segregate objects from its background is really an important task for robots to interact with surroundings in real life scenario. The advent of geometrically rich feature based objects has given impetus to research in the field of visual image recognition. This paper aims at providing a method for object detection with the help of SURF (Speeded-Up Robust Feature) algorithm used with KNN (K-Nearest Neighbors), a supervised Machine Learning Algorithm . This paper purports a method for object detection in which a set of features are extracted from an image captured from different perspectives. To increase the segmentation accuracy, we preform fixated point type segmentation on image and next key points are estimated in each segment with the help of Speeded-Up Robust Features (SURF). These key points are used to carry out the matching task for every detected key point in segmented image of the scene. We analyze and evaluate the performance of our system with other recently proposed methods such as the scheme pro-posed by Mae et al. Keywords:- Object Detection; K-NN; Segmentation; SURF; Feature Extraction; Voting. RESEARCH ARTICLE OPEN ACCESS