A METRIC BASED NOVEL CLASSIFICATION APPROACH TO KINECT POINT CLOUD DATA

Erdal ÖZBAY¹ , Ahmet ÇINAR²

Object classification in 3D point cloud data is an emerging topic attracting increasing research interest. Object detection is one of the most important challenges in computer vision. This paper proposes a novel method for the efficient detection of the real objects with respect to 3D point cloud data. The real object detection is performed using a technique based on mean shift clustering algorithm. The efficiency of the method is verified comparative official 3D data and real 3D point cloud data. We embed presented approach in a framework that combination of extracts shape and point cloud data metric information to improve the outcome of the classification stage. For this aim, classification of 3D point cloud data allows robust segmentation and feature descriptions into different objects by significantly reducing the error. Performed mean shift classification algorithm on the raw data and metric data classification with mean shift algorithm implementation are automatically compared to for evaluation the accuracy of the classification of metric classification algorithm. The results obtained metric classification algorithm and mean shift algorithm on automatic classification of simple planimetric object shapes with the surface of the point cloud show that proposed method is an efficient process.

Keywords
Classification, Point Cloud, Mean-Shift, Object Detection, Kinect,