A GPU based real-time image processing for an axis-symmetrical optical laser triangulation system

An axis-symmetrical optical laser triangulation system was developed by the authors to measure the inner geometry of long pipes used in the oil industry. It has a special optical configuration able to acquire shape information of the inner geometry of a section of a pipe from a single image frame. It uses a radial light sheet and conical triangulation to measure the inner geometry of pipes in cylindrical coordinates. A set of equally spaced images of 1024 x 1024 pixels is acquired at 50 Hz while the device is moved along the pipe’s axis. The measured geometry is analyzed to identify defects like corrosion damage. A GPU based processing algorithm has been developed to make the system able to process these images and display the geometrical/measurement result in real-time. The algorithm implements an adaptive threshold filter and a light intensity peak search using a graphic processing unit programming architecture (CUDA). Prior to the parallel algorithms (called kernels) a texture data type is used to remap the image, converting from polar to Cartesian coordinates, mapping angular and radial values in a 2D pixel data matrix. Radial lines are only scanned in a limited range (256 pixels) between a minimum and a maximum radius value. The achieved image processing frequency is about 470 frames per second (FPS) using a notebook equipped with a GTX 285m graphics card.