Robust Estimation Method Research On Spatial Attitude Based On Visual Feature Point Cloud

The mobile robot obtains its own spatial position and attitude information,which is usually done by photoelectric encoder and GPS.However,the photoelectric encoder is easily affected by the skidding of the tires,and the GPS can also cause the lack of signal in the complex environment.Compared with these traditional odometer techniques,visual odometer is more advantageous.Therefore,visual odometer plays an important role in robot motion control,SLAM and autonomous navigation.This paper focuses on the research of visual odometer technology,and proposes a better robust feature extraction matching and motion estimation algorithm,which provides a good foundation for real-time location and map building.First of all,based on computer vision research background,a comprehensive analysis of the current domestic and foreign visual odometry technology development history and current situation,and summed up the six steps of visual odometry,and expounds the theory of binocular stereo vision visual odometry,based on the camera calibration and distortion correction experiment.Secondly,in view of the commonly used feature extraction algorithm,SURF matching accuracy is low,calculation is too large,the matching efficiency is not high,puts forward an improved algorithm,feature rich SURF descriptor information and make them low dimension;while some nodes search with BBF tree,through the experiment,our improved algorithm can make the performance of the whole algorithm is improved,and its application to feature tracking in the image before and after.Finally,for the frequently used mathematical models of motion estimation,the least squares algorithm is prone to mismatch.The RANSAC algorithm is used to get more accurate models and outputs.The beam adjustment method is used to optimize the pose and get more accurate robust motion estimation.By collecting continuous image experimental data sets on campus,we get the motion estimation trajectory of mobile robot,and prove the authenticity and correctness of the proposed algorithm through comparison.