Research On 3D Scene Reconstruction Algorithms For Vision-based Navigation's Autonomous Land Vehicle

This paper describes my achievement during the graduated research in the field of Autonomous Land Vehicle (ALV), which is an intelligent mobile robot that can run autonomously and continuously on road or cross-country. Focuses on the specialties of the stereovision applied in vision-based navigation, we discussed a series of algorithms from error analysis and post-disposal to reconstruction of 3D scene from the disparity map.Chapter 3 presents a whole solution to reconstruct 3D scene from the disparity map. First, we discussed how to get the camera-based coordinates from the disparity map. Second, we developed the conversion from the camera-based coordinates to the vehicle-based one, and obtained much more practical reconstruction result. Finally, the quantity of data was reduced greatly by subdividing whole scene into smaller subunits.On the base of existing solution to reconstruct 3D scene from the disparity map, we analyzed various error sources and character of binocular stereo system on pinhole camera model for estimation of the depth. As a result, various errors are compensated. Thus, more practical and robust reconstruction result was obtained.The experiment results proved that our algorithm can meet the vision-based navigation's request for real-time and validity.