Abstract
Landmark extraction and matching is a key part of vSLAM. In this paper, a method of landmark extraction and its description based on three-dimensional (3D) information of feature points is proposed. The robot obtains images from its environment via binocular vision, building 3D information of feature points under left camera coordinate system. Then, it uses the method based on improved mean shift algorithm to extract natural landmarks and construct its descriptor, which can be used in fast matching of two landmarks. This method can extract natural landmarks from unstructured environments and tolerate relatively low accuracy of pose estimation. It is suitable for navigation system of a mobile robot running in a complex environment. vSLAM experiments showed its effectiveness.
Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
