Research On Object Deformation Quantity Extraction In Three-dimension Scene Reconstruction

Due to the imaging principle and the inherent optical characterist ics of the imaging equipment, the Kinect depth image has various errors, which will limit the extraction of high accuracy depth value. In order to improve the qualit y o f depth image for obtaining high quality depth informat ion and deformatio n quantity, the main research work is as fo llows.(1) An enhancement algorithm for single depth image. After analyzing the principle of Kinect depth distance measurement, it is found that there are three main problems: unmatched edges between depth image and its corresponding color image, ho les caused by invalid pixels, the inherent error of the depth value. In order to solve these problems, firstly, the unmatched pixels in the edge of the two images are detected by the method of region growing and the pixel value is set to 0. Then, the method of combining the median filter and the jo int bilateral filter is proposed to fill the hole in the image. Finally, the bilateral filter is used to remove the image noise, which can eliminate the error of the depth value.(2) An enhancement algorithm for depth sequence. This improved algorithm is based on single depth image enhancement algorithm. The specific method is to establish a relat ionship between various d epth images by Kalman filter algorithm. Then the depth image sequences are optimally estimated to obtain the optimal depth image. Finally, the optimal depth image is denoised to eliminate flicking in the time domain, and the deformation quantity is extracted through the depth values from different distances.(3) After getting and enhancing the depth image sequence of different distance, the depth informat ion of different distance can be used to realize the extraction of deformation quantity.