Recover 3D Information of the Moving Object from Video Streams

Abstract

Perception of the moving object in 3D from video streams has been one hot topic in computer vision. We present a fast method to reconstruct 3D information of the moving object from binocular video streams. System is assembled as two pipelines, technica are used to excavate the potential parallelism. With the corresponding points searching confined to very limited and credible region, the mismatching errors and time-consumed computation are reduced considerably. At the last, sparse depth map is calculated and then 3D contours and location of the object are estimated. The system is implemented and tested with outdoor and indoor moving object perception on 640×480 frame. Results show that the proposed method is improved in speed and stability. It can be used as a reference for autonomous navigation of mobile robot and object tracking.

Original document

The different versions of the original document can be found in:

• http://dx.doi.org/10.2991/icmt-13.2013.202 under the license https://creativecommons.org/licenses/by

• https://download.atlantis-press.com/article/10554.pdf

• https://www.atlantis-press.com/proceedings/icmt-13/10554,

https://academic.microsoft.com/#/detail/2042859991 under the license cc-by