3D point cloud generation from 2D depth camera images using successive triangulation

Depth cameras are low-cost, plug & play solution to generate point cloud. 3D depth camera yields depth images which do not convey the actual distance. A 3D camera driver does not support raw depth data output, these are usually filtered and calibrated as per the sensor specifications and hence a method is required to map every pixel back to its original point in 3D space. This paper demonstrates the method to triangulate a pixel from the 2D depth image back to its actual position in 3D space. Further this method illustrates the independence of this mapping operation, which facilitates parallel computing. Triangulation method and ratios between the pixel positions and camera parameters are used to estimate the true position in 3D space. The algorithm performance can be increased by 70% by the usage of TPL libraries. This performance differs from processor to processor.