The Research Of Omni-directional Intelligent Wheelchair Motion Estimation Method Based On Variational Optical Flow

With the aged tendency of population and the need for improving the quality of life for the disabled, to enhance the automation level of intelligent wheelchair independent new technology has become an urgent research subject research. Intelligent omnidirectional movement wheelchair with mecanum wheel has good flexibility of the movement, but it challenges velocity measurement of the omnidirectional wheelchair.As the traditional velocity estimation method based on optical flow does not perform robustly against outliers caused by non-uniform ambient illuminations, motion blurring and obstacles, and the accuracy of motion estimation is affected when the wheelchair velocity is larger. To solve these problems, an improved variational optical flow based technique to perceive the motion velocity of an omnidirectional smart wheelchair robotics was proposed. The method was verified by experiments conducted on Pioneer3 robot and omnidirectional intelligent wheelchair with cameras system. The research work mainly include the following aspects:(1) The improved variational optical flow estimation methodThe traditional variational optical flow computation model is sensitive to the constant changes of illumination and may cause over-smoothing in the discontinuities of flow fields and image. To solve these problems, the advanced data terms including the gradient value constancy assumptions and the laplacian constancy assumptions, which are insensitivity to changes of illumination and suitable for non-translational displacements, are introduced in this paper; An estimation technology of variational optical flow based on image and flow-driven regularization methods are introduced to preserve the discontinuities of flow fields and image; Finally, the algorithm is implemented under the framework of CUDA to improve the real-time performance of the algorithm.(2) RANSAC refinement based on planar surface optical flow modelIn indoor scene, it is hard to avoid that performance of flow fields deteriorate in the presence of non-uniform ambient illuminations, local motion blur and obstacles, it will contribute to erroneous motion estimations. To solve these problems, parameterized optical flow methods assume the image motion field of a moving planar surface can be expressed as a quadratic polynomial of the image coordinates. To reject invalid velocity, we therefore attempt to estimate the quadratic model in the framework of Random Sample Consensus(RANSAC) algorithm. The planar surface optical flow model based RANSAC method can remove outlier vectors produced by non-uniform ambient illumination, local motion blur and obstacles.(3) Kalman Filter(KF)-integrated omnidirectional intelligent wheelchair velocity estimationDue to mechanical vibrations, motion shock, and sometimes imprecision of optical flow vector estimations, the recovered velocity appears to be oscillational, by virtue of KF, the oscillation in posterior estimation is relieved dramatically. The faster wheelchair travels, the less image overlap for accurate velocity estimation. To overcome the problem of dealing with large image displacements, KF is incorporated to efficiently predict the image transformations. Reducing the feature search area, the KF enables the variational optical flow method to rapidly converge and give accurate velocity estimates. As demonstrated from experimental results that, the accuracy of omnidirectional wheelchair velocity estimation can be improved by estimating dense variational optical flow.