3D points recover from stereo video sequences based on open CV 2.1 libraries

Abstract

The purpose of this study was to implement a program in C++ using OpenCV image processing platform's algorithms and Microsoft Visual Studio 2008 development environment to perform cameras calibration and calibration parameters optimization, stereo rectification, stereo correspondence and recover sets of 3D points from a pair of synchronized video sequences obtained from a stereo configuration. The study utilized two pretest laboratory sessions and one intervention laboratory session. Measurements included setting different stereo configurations with two Phantom v9.1 high-speed cameras to: capture video sequences of a MELFA RV-2AJ robot executing a simple 3D path, and additionally capture video sequences of a planar calibration object, being moved by a person, to calibrate each stereo configuration. Significant improvements were made from pretest to intervention laboratory session on minimizing procedures errors and choosing the best camera capture settings. Cameras intrinsic and extrinsic parameters, stereo relations, and disparity-to-depth matrix were better estimated for the last measurements and the comparison between the obtained sets of 3D points (3D path) with the robot's 3D path proved to be similar.