Identification and correction of flying pixels in range camera data

This paper focuses on errors which are possible to occur in the depth measurements of range cameras. Range cameras can capture 3D information of a scene by sending out infrared light and then measuring the reflections. Wrong measurements occur at the edges of objects where the depth level changes. A depth value between the foreground and background level is measured which creates a so-called "flying pixel" when displaying the 3D points. In this paper different methods for the identification of flying pixels are presented and compared. The advantages and drawbacks of each method are discussed. Then a simple method for the correction of flying pixel errors is presented and its limitations are shown. The final method for correction is presented which is based on segmenting the pixel matrix into horizontal and vertical scanlines. After segmentation, linear segments can be identified to which the pixels can be mapped. The paper concludes with the evaluation of the presented methods to show their effectiveness.