Design And Implementation Of Quadrotors Based On Stereo Vision

For the design and the implementation of a visual SLAM system which can be applied to the quadrotors,this paper studies and tests the existing open source visual SLAM system in-depth.The disadvantages of the existing typical systems are summed up and improved.For real-time purpose,a stereo vision SLAM system based on semi-direct method is designed and implemented based on the feature point method and the semi-direct method.In order to improve the robustness of the system,a binocular vision sensor is used to estimate the three-dimensional space pose estimation.The estimation of the real environment scale information can thus be realized by stereo matching algorithm.The validity of the algorithm is verified by experiments.The main contributions of this paper can be summarized as follows:Firstly,the position model of quadrotor and the basic model of the visual system are built.By using probability distribution model,the relevant coordinate system is defined and the system-related model is established.The geometric principles of the visual SLAM system are analyzed using the camera model.The difficulties of the application of the visual SLAM system to quadrotors are summarized and the foundation for the system design and improvement is laid.Secondly,the overall system framework is proposed.The actual requirements and the indexes of the system are determined by previous research and tests.According to the motion estimation method,the mainstream visual SLAM systems can be classified into the filter based method,the keyframe BA based method and the direct method.The corresponding existing advanced representative systems are tested according to the classification results.The advantages and disadvantages of these typical systems are compared.Taking real-time performance and robustness into account,the overall system framework is ultimately determined.Finally,the stereo vision matching algorithm for quadrotor based on semi-direct method is proposed.Considering the large computation and poor robustness under strong rotation of the ORB-SLAM system based on the feature point method and non-estimation of real environment scale of the SVO system based on semi-direct method,this system takes both real-time performance and robustness into account.Referring to the idea of SVO system,the semi-direct feature is used for the stereo feature matching.The stereo vision SLAM system based on semi-direct method is realized with high computational efficiency and robustness.Finally,the feasibility of the proposed system is verified by corresponding experiments.