Dual-Side Monocular Visual Odometrv System Based On Smart Mobile Devices

Smart mobile devices have been the must have part of most people’s life.Therefore,lots of location-ware applications have been developed based on that.The devices can take advantage of some satellite system such as GPS,GLONASS to locate themselves with high accuracy outdoor.However,the inertance system and the visual odometry system are almost the only ways to locate and navigate without external signal indoor.But the inertance system’s accuracy after long time running can barely satisfy the application and it can hardly be improved due to the limitation of sensors,so that the visual odometry is the only ideal way to do that.Traditional visual odometry system is designed for some large platforms such as unmanned motor and this paper pay more attention to the visual odometry which run on mobile platforms that have limited computing power and sensors.Some previous work on the homogeneous platforms mostly focused on the performance of accuracy and real time capability,there’s high standard of the environment system running in.Moreover,the smart mobile devices mostly have the back and front cameras can work simultaneously,this is different from the traditional monocular visual odometry and the stereo visual odometry which required the two cameras have shared field of view.This paper focus on the simultaneously monocular visual odometry with back and front cameras based on the previous work about monocular visual odometry,real time visual odometry with the hardware of smart mobile devices combined.The main work and contribution in this paper shows below:(1)Propose the design of the visual odometry system with back and front camera working simultaneously.Setting up the software and hardware platform of the system and introduce the alignment of space and time.(2)Design and accomplish the comparison experiments to evaluate the accuracy and the performance of three mainstream feature detectors algorithms:SIFT,ORB and FAST in the real indoor environment.Then achieve the monocular visual odometry system based on the SVO algorithm and the experiments’ result.(3)Propose the algorithms to evaluate the status of running single-side visual odometry by the covariance matrixes and the estimated speed to the situations that one side’s failure.Propose one algorithm to restart and resume the single-side visual odometry to avoid the jump of system’s output.Then this paper designs and does some experiments about this situation.The result validates the algorithm can solve the problems and really improve the robust of visual odometry system.(4)To the situation dual side run normally,the paper proposes one algorithm to fusion the output of both side using the kalman filter.The paper explains the model used and makes some experiments to check the validity.By the error analysis of the experiments,the paper showed that the algorithm improves the accuracy clearly.From the experiments and the result,the dual-side visual odometry system the paper proposed can improve the robust of visual odometry on the smart mobile devices when the environment is severe and the accuracy when the environment is ideal indoor.