Depth estimation from 3D reconstructed scene using stereo vision

• Main Author: Omer Ba-saleem, Omer Mohamed
• Format: Thesis
• Language: English; English
• Published: 2018
• Subjects: TA1501-1820 Applied optics. Photonics
• Online Access: http://eprints.uthm.edu.my/192/

In the past few years, a growing interest in mobile robot’s applications produce a persistent need to develop assistive systems that can help those robots to do their job efficiently. Distance estimation is one of these applications that a lot of researches made to develop it. Visionary sensors such as stereo vision optimization method is one of the most efficient methods among traditional sensors that have been used to estimate a depth from 3D reconstructed scene. Vision sensor can be used in rescue robots mission instead of human providing a way to estimate the distance between the robot camera and the objects in front it and give a 3D model of that obstacle which can help rescuers. Although vision sensors have some disadvantages such as the change of illumination of the captured images and the change in the background. In this project, the objectives are developing a stereo vision system to extract depth from 3D reconstructed scene, analyse the performance of the proposed stereo vision system and develop a graphical user interface (GUI) to monitor the performance of the stereo vision system. Two low-cost cameras have been used to works as stereo camera sensor. Stereo camera parameters from the camera calibration process and disparity map by using the block matching in order to collect the corresponding pixels between both stereo images are important parameters to estimate the depth from the 3D reconstructed scene. The experiment results have been done by testing an object far from the stereo camera by 1.0 and 1.5 meters and the experiment conducted at indoor and outdoor environments. The results show that the mean error of depth increasers with increasing distance to the stereo camera. In addition, the quality of the texture of the 3D virtual model also decreased with the increasing of the distance. The camera mean accuracy reaches to 90.14%. Finally, using stereo vision in order to estimate depth from the scene have been done and it can be use at rescue missions to help at rescue’s robots missions.