Research Methods For Visual Odometry Based On The Compound Model Camera

Visual odometry (VO) is the process of estimating the egomotion of an agent(e.g.,vehicle,human and robot) using only the input of a single or multiple camerasattached to it, which can provide position information to autonomous robotpositioning, creating map and navigation.Through the movement of image information, visual odometry can estimate theposition changes of an object, and there aren’t data errors causing by lower sensorprecision and inertial navigation drift (loose soil, ground skid). As wheeled odometrycannot satisfy the legged robot navigation and positioning, IMU inertial navigationcannot apply in long distance navigation positioning, GPS global satellite positioningsystem can’t satisfy the underwater, polar and interstellar detection and other fields,visual odometer has an irreplaceable advantage.RGB-D camera based on ommateum model is a kind of new3D vision system,which can not only obtain2D color images of the scenario but also the correspondingthree-dimensional information according to the pixel. Based on RGB-D camerafeatures, this paper puts forward a kind of high precision three-dimensional motionadaptive robust estimation method based on image features and the three-dimensionalscene information.This method use the sparse of scene image features as feedback quantity to adjustthe sampling time of adjacent frames adaptively, and switch three dimensionalestimate optimization scheme automatically. Reducing the sampling frequency of therich image feature scene, obtaining accurate3d motion estimation by extracting SIFTfeature matching point sets and the corresponding three-dimensional information,which can reduce the acquisition of processing and improve the reliability of motionestimation; Increasing the sampling frequency of sparse image features scene,limiting the movement of adjacent frame in a small space, using3D point cloudinformation of adjacent frames directly, using ICP iterative algorithm can solve the3D motion estimation reliably, which can reduce the dependence on density ofEnvironmental image features effectively. To reduce influence of the error matchingimage features of3D estimates, this method introduced consistency check constraints,color consistency constraints and RANSAC consistency constraints, which caneliminate false matching feature points on image effectively.Compared with the existing visual odometry3D motion estimation methods, this scheme has great advantages such as high precision, high reliability, small amount ofcalculation and fewer restrictions of application scenario and so on. The experimentalresults verify the feasibility and effectiveness of the proposed method.