THREE PRE-PROCESSING STEPS TO INCREASE THE QUALITY OF KINECT RANGE DATA
Keywords: Accuracy, Registration, Point Cloud, Edge Detection, Close range, Range-Image, Kinect
Abstract. By developing technology with current rate, and increase in usage of active sensors in Close-Range Photogrammetry and Computer Vision, Range Images are the main extra data which has been added to the collection of present ones. Though main output of these data is point cloud, Range Images themselves can be considered important pieces of information. Being a bridge between 2D and 3D data enables it to hold unique and important attributes. There are 3 following properties that are taken advantage of in this study. First attribute to be considered is "Neighborhood of Null pixels" which will add a new field about accuracy of parameters into point cloud. This new field can be used later for data registration and integration. When there is a conflict between points of different stations we can abandon those with lower accuracy field. Next, polynomial fitting to known plane regions is applied. This step can help to soften final point cloud and just applies to some applications. Classification and region tracking in a series of images is needed for this process to be applicable. Finally, there is break-line created by errors of data transfer software. The break-line is caused by loss of some pixels in data transfer and store, and Image will shift along break-line. This error occurs usually when camera moves fast and processor can't handle transfer process entirely. The proposed method performs based on Edge Detection where horizontal lines are used to recognize break-line and near-vertical lines are used to determine shift value.