SLAM Algorithm Based On Bumblebee XB3

Accurate 3D scene informention is an important facter in autonomous driving and mobile system navigation. Intelligent robots often make their decisions according to observations obtaining from stereo cameras which are cheap and easy to employ. The complexity of real environment around the robots, the inevitable noise of observation data and even the missing information of sensors make visual SLAM more difficult to carry out. Although it has already been demonstrated successfully in both Highway navigation and autonomous driving in simple environment, autonomous navigation in unknown environment remains an unsolved problem. This paper researches a approach which is used to estimate the egomotion of the robot and features’ 3D informention from a sequence of stereo images. All data is provided by a stereo camera called Bumblebee XB3 which is mounted on top of the intelligent robot platform.This paper mainly studies binocular visual SLAM which estimates the egomotion of the robot and features’ 3D informention. Accurate 3D information from video sequences is a core subject in computer vision and robotics. Rough features are achieved by matching features in a ’circle’. This algorithm can reduce computation time, and accelerate the matching speed. And we apply Random Sample Consensus (RANSAC) algorithm and epipolar geometric constraint to eliminate false matching points.We estimate the egomotion of the robot by using EKF-SLAM algorithm base on accurate matching features. The experiment verifies the effectiveness of the binocular visual SLAM algorithm in the outdoor unknown unstructured situations.We employ matching algorithm in conjunction with the binocular visual SLAM algorithm to obtain dense features. According to the estimated robot motion, we project the dense features into a robot coordinate system to build accurate 3d map.