Monocular Vision SLAM Algorithm Based On PCA-SURF

With the rapid development of science and technology, computing speed andsensor technology, the autonomous and intelligent technology of mobile robot hasalso been rapid development. One of the robot’s tasks is to identify and locate target,visual sensor has been widely used in mobile robots because of its advantages of largeamount of information. In order to achieve the autonomy of mobile robot, two of themost fundamental questions are environment map building and autonomouslocalization. In an unknown environment, mobile robots need to carry sensors toperceive their environment information surrounding through observation environment,incremental environment map construction, autonomous positioning and use the mapinformation to achieve autonomous robot that is the Simultaneous Localization andMapping (SLAM). Comparing with the sonar and laser range finder, visual sensor hasthe advantages of cheap price, small size, high versatility, and more abundant for theperception of environmental information, therefore it is applied more and more widelyin the mobile robot.This paper mainly studies the monocular visual SLAM, which has theadvantages of simple calculation, wide range of application relative to stereo visionand the panorama visual SLAM, and also can be combined with the odometer toachiever the three-dimensional measurement information on the surroundingenvironment, flexible and less distortion.Image feature points extraction is an important step of monocular visual SLAM,requiring a rapid detection method and robustness. This paper proposes a PrincipalComponent Analysis-Speeded Up Robust Feature (PCA-SURF) feature extractionalgorithm, which has advantages including a small amount of calculation, geometricdistortion, high robustness, etc. When extracting feature points, using the method ofextracting a certain amount of random features can make the features evenlydistributed in the image, reduce the amount of calculation of image matching, andaccelerate the matching speed. Using one point Random Sample Consensus (RANSAC) algorithm to eliminate false matching points.This paper further describes the monocular visual SLAM system structure basedon the Extended Kalman Filter (EKF), integrated PCA-SURF algorithm as well asthe map features of inverse depth parameter and one point RANSAC (Random SampleConsensus) algorithm, also integrated the comprehensive design experiment, andverifies the effectiveness of the monocular visual SLAM algorithm in the laboratoryunknown unstructured situations.Through experimental verification, compared to the monocular visual SLAMalgorithm based on fast Harris corner, the monocular visual SLAM algorithm basedon PCA-SURF has a stronger robustness, better real-time processing effect, precisepositioning of camera, and build real-time monocular visual SLAM system.