Research For The Visual Odometry Optimized Algorithms Based On The Three-dimensional Camera

Visual odometry is the process of estimating the egomotion of an agent which using the associated single or multiple cameras to obtain the images and calculating the relationship between them,and itcan provide position information to the visial navigation,wearable computing,3-D map creation.The absolutely scale information is missing in the monocular visual odometry, thus it requires the fuzziness scale factor in the process of solving the model and needs the certain constraints and assumptions.The binocular visual odometry can provide the depth information based on the fixed baseline,but it is similar to the monocular odometry when the object distance exceed a certain range,and it must be used in the appropriate range.The three-dimensional visual odometry can dirtectly obtain the depth information,so it could extract the feature and calculate the motion estimation using the depth and color picture information.This paper proposes anoptimization algorithm for the visual odometry program based on three-dimensional camera Kinect, which could slovethe problemsthat existsin the development of visual odometey.Firstly, the optimization algorithm uses Kinect camera to obtain the depth and color picture information,it could ensure the correspondence between the depth and color images in the image acquisition process;Then it use the ORB(Oriented FAST and Rotated BRIEF) algorithm to extract and match feature points, it couldensure the speed and accuracy of program;Then it used RANSAC(Random Sample Consensus) algorithm to culling mismatching points, which could ensure the accuracy and reliabilityof the match points and lay a good foundation for the latter part of the motion estimation;Next we take the matching feature pointsindex to the corresponding three-dimensional information to obtain accurate three-dimensional matching feature points, and then use 3D-RANSAC algorithm and ICP(Iterative Closest Point) algorithm to get the motion estimation's rotation and translation matrix, which could get reliable motion estimation both the sufficient feature points and the shortage feature points; Finally, the SBA(Sparse Bundle Adjustment) global optimization algorithm is used to achieve overall optimization of the motion estimation, reducing the accumulated error ofthe motion estimation calculate process, it could ensure the entire motion estimation's accuracy.This algorithm has good accuracy, reliability, computati onal speed, and has a wider range of applications, compared with the currently available methods for solving the visual odometey. Experimental results demonstrate the accuracy and reliability of the algorithm.