Motion Estimation Based on an Improved Optical Flow Method Using PIV for VSLAM

As an important research topic in the field of robotics, the optical flow method is widely used for motion estimation in visual simultaneous location and mapping (VSLAM). However, the optical flow method is based on the gray-invariant assumption, which restricts its application in the case of drastic luminosity variation. Moreover, the optical flow method can speed up the processing speed in motion estimation, but it cannot work effectively in scenarios with missing features. As a velocity measurement technology in the field of flow field and fluid, the particle image velocimetry (PIV) can overcome the aforementioned disadvantages of the optical flow method, and achieve motion estimation since it considers points in an image uniformly and can achieve sub-pixel accuracy for positional estimation. To this end, an improved optical flow method based on PIV is proposed by adopting the FFT cross-correlation matching algorithm and the sub-pixel displacement matching algorithm to estimate the image pixel displacement in the field of missing features. Experiments on the EUROC data-set show that the proposed method can not only track the motion of more pixels compared with that the multi-layer optical flow method, but also run in higher accuracy in the areas with missing features.